OpenFog Reference Architecture 2 09 17 PDF

OpenFog Reference Architecture
for Fog Computing
Produced by the OpenFog Consortium Architecture Working Group

www.OpenFogConsortium.org
February 2017
1
OPFRA001.020817 © OpenFog Consortium. All rights reserved.
Use of this Document
Copyright © 2017 OpenFog Consortium. All rights reserved. Published in the USA.
Published February 2017.
This is an OpenFog Consortium document and is to be used in accordance with the

terms and conditions set forth below. The information contained in this document is
subject to change without notice.
The information in this publication was developed under the OpenFog Consortium
Intellectual Property Rights policy and is provided as is. OpenFog Consortium
makes no representations or warranties of any kind with respect to the information
in this publication, and specifically disclaims implied warranties of fitness for a
particular purpose. This document contains content that is protected by
copyright. Copying or distributing the content from this document without
permission is prohibited.
OpenFog Consortium and the OpenFog Consortium logo are registered trademarks
of OpenFog Consortium in the United States and other countries. All other
trademarks used herein are the property of their respective owners.
Acknowledgements
The OpenFog Reference Architecture is the product of the OpenFog Architecture
Workgroup, co-chaired by Charles Byers (Cisco) and Robert Swanson (Intel). It
represents the collaborative work of the global membership of the OpenFog
Consortium. We wish to thank these organizations for contributing to this work and
to the advancement of fog computing technology, research and innovation: Aalto
University; ABBALab; Arizona State University; ARM; AT&T; Caltech; Cisco; Dell;
FogHorn Systems; Fujitsu; GE Digital; Hitachi; Foxconn; Indian Institute of
Technology; Industrial Technology Research Institute; Institute for Information
Industry; Institute of Network Coding; The Chinese University of Hong Kong; Intel;
Internet Initiative Japan Inc.; ITOCHU techno-Solutions Corporation; Kii; LGS
Innovations; MARSEC; Microsoft; Mitsubishi Electric Corporation; National Chiao
Tung University; National Taiwan University; Nebbiolo Technologies; NEC
Corporation; NGD Systems; NTT Communications; OSIsoft; Princeton University;
PrismTech; Real-Time Innovations; relayr; SAKURA Internet; Schneider Electric;
Shanghai Institute of Microsystem and Information Technology; ShanghaiTech
University; Singapore University of Technology and Design; SRC Inc.; Stichting
imec Nederland; The Chinese University of Hong Kong; Toshiba; Technische
Universität Dresden; TTTech; University of Colorado Boulder; University of Georgia;
University of Pisa; University of Southern California; Vanderbilt University; Wayne
State University.
2
OpenFog Overview
Digital innovation from the Internet of Things (IoT), Artificial Intelligence,
Virtual Reality, Tactile Internet and 5G applications is transforming the way
we work, commute, shop and play. Data from newly-connected factories,
homes, communities, cars, hospitals and more is expected to grow from 1.1
zettabytes (or 89 exabytes) per year in 2016 to 2.3 zettabytes (or 194
exabytes) per year by 2020.1 Current “cloud-only” architectures cannot keep
up with the volume and velocity of this data across the network, thereby
reducing the value that can be created and captured from these
investments.
Fog computing provides the missing link in the cloud-to-thing continuum.

Fog architectures selectively move compute, storage, communication,
control, and decision making closer to the network edge where data is being
generated in order solve the limitations in current infrastructure to enable
mission-critical, data-dense use cases.
Fog computing is a:
A horizontal, system-level architecture that distributes

computing, storage, control and networking functions closer to
the users along a cloud-to-thing continuum.
Fog computing is an extension of the traditional cloud-based computing

model where implementations of the architecture can reside in multiple
layers of a network’s topology. However, all the benefits of cloud should be
preserved with these extensions to fog, including containerization,
virtualization, orchestration, manageability, and efficiency. In many cases,
fog computing works with cloud. Pillars, which are common themes of the
OpenFog reference architecture include security, scalability, openness,
autonomy, RAS (reliability, availability and serviceability), agility, hierarchy,
and programmability. In addition to the pillars, we describe the roles of each
stakeholder in the fog value chain from silicon creator to the Operating
System and application developer through a composite architectural
description
Fog computing also is often erroneously called edge computing, but there
are key differences. Fog works with the cloud, whereas edge is defined by
the exclusion of cloud. Fog is hierarchical, where edge tends to be limited to
a small number of layers. In additional to computation, fog also addresses
networking, storage, control and acceleration.
3
The OpenFog Consortium was formed on the principle that an open fog
computing architecture is necessary in today’s increasingly connected world.
Through an independently run open membership ecosystem of industry, end
users and universities, we can apply a broad coalition of knowledge to these
technical and market challenges. We believe that proprietary or single
vendor solutions can limit supplier diversity and ecosystems, resulting in a
detrimental impact on market adoption, system cost, quality and innovation.
It is our intent to ensure the OpenFog reference architecture results in fully
interoperable and secure systems, supported by a vibrant supplier
ecosystem.
4
Contents
Use of this Document ........................................................................................................................... 2
Acknowledgements ................................................................................................................................ 2
OpenFog Overview ................................................................................................................................. 3
1 About Fog Computing and the Consortium ........................................................................................... 9
OpenFog Reference Architecture Overview.................................................................................... 9
2 Areas of Opportunity ............................................................................................................................ 10
OpenFog Reference Architecture Content .................................................................................... 10
The Internet of Things, Cloud and the OpenFog RA ...................................................................... 10
OpenFog and Other Consortia ...................................................................................................... 12
3 Use Cases for Fog ................................................................................................................................. 13
Transportation Scenario: Smart Cars and Traffic Control ............................................................. 14
Visual Security and Surveillance Scenario ..................................................................................... 17
Smart Cities Scenario ..................................................................................................................... 19
3.3.1 Smart Buildings ........................................................................................................................... 20
Additional Use Cases ..................................................................................................................... 21
4 Pillars of OpenFog RA ........................................................................................................................... 22
Security Pillar ................................................................................................................................. 23
Scalability Pillar.............................................................................................................................. 24
Openness Pillar .............................................................................................................................. 26
Autonomy Pillar ............................................................................................................................. 27
Programmability Pillar ................................................................................................................... 29
Reliability, Availability, and Serviceability (RAS) Pillar .................................................................. 29
Agility Pillar .................................................................................................................................... 31
Hierarchy Pillar .............................................................................................................................. 31
4.8.1 Hierarchical Fog Deployment Models ......................................................................................... 34
5 Reference Architecture Overview ........................................................................................................ 38
Functional Viewpoint .................................................................................................................... 38
Deployment Viewpoint ................................................................................................................. 39
5.2.1 OpenFog Deployment Types ....................................................................................................... 39
5.2.2 N-Tier Fog Deployment ............................................................................................................... 39
5
OpenFog Architecture Description ................................................................................................ 43
Perspectives (Cross Cutting Concerns) .......................................................................................... 46
5.4.1 Performance and Scale Perspective............................................................................................ 46
5.4.2 Security Perspective .................................................................................................................... 47
5.4.3 Manageability Perspective .......................................................................................................... 51
5.4.4 Data, Analytics, and Control ....................................................................................................... 54
5.4.5 IT Business and Cross-fog Applications ....................................................................................... 56
Node View ..................................................................................................................................... 56
5.5.1 Network ...................................................................................................................................... 58
5.5.2 Accelerators ................................................................................................................................ 65
5.5.3 Compute...................................................................................................................................... 66
5.5.4 Storage ........................................................................................................................................ 67
5.5.5 OpenFog Node Management...................................................................................................... 68
5.5.6 OpenFog Node Security .............................................................................................................. 69
System Architecture View ............................................................................................................. 73
5.6.1 Hardware Platform Infrastructure .............................................................................................. 74
5.6.2 Hardware Virtualization and Containers .................................................................................... 77
Software Architecture View .......................................................................................................... 77
5.7.1 Software View Layers .................................................................................................................. 77
6 Adherence to OpenFog Reference Architecture .................................................................................. 89
7 An End-to-End Deployment Use Case .................................................................................................. 91
Airport Visual Security ................................................................................................................... 91
7.1.1 Cloud and Edge Approaches ....................................................................................................... 92
7.1.2 Fog Computing Approach ........................................................................................................... 93
7.1.3 Application to Airport Visual Security ....................................................................................... 103
8 Additional Opportunities .................................................................................................................... 120
9 Summary and Next Steps ................................................................................................................... 121
10 Appendix – Deeper Security Analysis ................................................................................................. 122
Security Aspects ........................................................................................................................ 122
10.1.1 Cryptographic Functions ........................................................................................................... 122
10.1.2 Node Security Aspect ................................................................................................................ 127
6
10.1.3 Network Security Aspect........................................................................................................... 130
10.1.4 Data Security Aspect ................................................................................................................. 137
11 Glossary .............................................................................................................................................. 143
References ........................................................................................................................................ 159
Figures
Figure 1 Unfettered Cloud Computing .................................................... 11
Figure 2 OpenFog Consortium and Other Consortia ................................. 12
Figure 3 OpenFog Transportation: Smart Car and Traffic Control System ... 15
Figure 4 Opportunities for Smart Cities .................................................. 19
Figure 5 Pillars of OpenFog .................................................................. 22
Figure 6 Layered Architecture View of an IoT System .............................. 32
Figure 7 IoT System Deployment Models ............................................... 34
Figure 8 Fog Hierarchy Example ........................................................... 36
Figure 9 Fog Hierarchical Deployment Model .......................................... 36
Figure 10 Multi-Tier Deployment ........................................................... 39
Figure 11 Intelligence from data ........................................................... 41
Figure 12 Fog Node East/West Communication ....................................... 43
Figure 13 Architecture Description with Perspectives ............................... 44
Figure 14 OpenFog Architecture Perspectives ......................................... 46
Figure 15 OpenFog Security Layers ....................................................... 48
Figure 16 Example Threats and Attacks ................................................. 49
Figure 17 Management Lifecycle ........................................................... 52
Figure 18 Management Layer ............................................................... 54
Figure 19 Business Intelligence ............................................................ 55
Figure 20 Cross Fog Application ............................................................ 56
Figure 21 Node View ........................................................................... 57
Figure 22 System Architecture View ...................................................... 74
Figure 23 Software Architecture View .................................................... 78
Figure 24 Application Support .............................................................. 82
Figure 25 Containerization for Application Support .................................. 82
Figure 26 Application Services Layer ..................................................... 85
Figure 27 Containerization for Application Support .................................. 86
Figure 44 OpenFog Technology Ready ................................................... 89
Figure 43 OpenFog Ready .................................................................... 89
Figure 28 Airport Scenario ................................................................... 91
Figure 29 Key pillars of the OpenFog Architecture ................................... 93
Figure 30 OpenFog Architecture Description ........................................... 94
7
Figure 31 Node view for Visual Security ................................................. 95
Figure 32 OpenFog Approach to Visual Security Scenario ....................... 104
Figure 33 OpenFog realized for Visual Security ..................................... 105
Figure 34 Training and Classification system for Machine Vision .............. 106
Figure 35 Airport License Plate Capture ............................................... 107
Figure 36 Integrated Air Travel Infrastructure - Assumptions ................. 108
Figure 37 Garage Gate Entrance ......................................................... 109
Figure 38 Central System Analytics ..................................................... 110
Figure 39 Terminal Entrance .............................................................. 111
Figure 40 Baggage flow ..................................................................... 113
Figure 41 Departure Gate .................................................................. 116
Figure 42 Arrival Gate ....................................................................... 117
Figure 50 OpenFog Node Security Architecture ..................................... 127
Figure 45 OpenFog Security Functional Layers and Operational Planes .... 131
Figure 46 OpenFog Secure Communication Pathways ............................ 132
Figure 47 Protocol Suites for Secure Node-to-Node Communications ....... 133
Figure 48 Protocol Suites for Secure Node-to-Node Communications ....... 133
Figure 49 Protocol Suites for Secure Node-to-Device Communications ..... 135
8
1 About Fog Computing and the
Consortium
OpenFog Reference Architecture Overview

The OpenFog Consortium was founded by ARM, Cisco, Dell, Intel, Microsoft
and Princeton University in November 2015. Through its global membership
of leading technology & networking players, fog computing entrepreneurs
and university researchers, the Consortium is helping to enable game-
changing innovation enabled by fog computing through an open architectural
framework.
We are guided by the OpenFog Board of Directors and the OpenFog

Technical Committee. The technical committee is the technical governing
body of all of the working groups of the Consortium. The chair of this group
is elected by a vote of the OpenFog Board of Directors and reports directly to
the board.
The OpenFog Reference Architecture (OpenFog RA) is intended to help

business leaders, software developers, silicon architects, and system
designers create and maintain the hardware, software and system elements
necessary for fog computing.
Many different technical workgroups in the Consortium are responsible for

the different aspects of this reference architecture, including:
Communications, Software-Infrastructure and Security. The Architecture
Framework workgroup is tasked with the creation and maintenance of this
document and other technical publications of the Consortium. All new
technical topics requiring investigation are assigned to this group. The
charter of each group is managed and approved by the technical committee
and Board of Directors.
For further information on these groups or how to participate, please

reference www.openfogconsortium.org.
9
2 Areas of Opportunity
OpenFog Reference Architecture Content
The OpenFog RA is part of a suite of documents under construction by the

OpenFog Consortium and our technical liaisons partners. It is a medium- to
high-level view of system architectures for fog nodes and networks. Future
documents will provide lower-level details, including formal, enumerated
requirements that will form the basis of quantitative testbeds and the
specified interoperability of fog elements. Future documents will also refine
the use cases described in Chapter 3.
The OpenFog RA is divided into the following chapters:

 Chapter 2 describes the OpenFog Consortium’s mission and plans to
accelerate fog computing. It also provides an overview of the OpenFog
RA.
 Chapter 3 presents some use cases where we see fog computing
emerging. This list will grow and evolve as the OpenFog RA is refined.
 Chapter 4 describes the pillars of the OpenFog architecture. These are
the guiding principles for the OpenFog RA.
 Chapter 5 provides an in-depth look at the full OpenFog RA. This
section shows the abstract Architectural Description (AD) for the
OpenFog RA.
 Chapter 6 starts the conversation on adherence to the OpenFog
architecture. It is our intention to drive standardization across the
various interfaces.
 Chapter 7 shows the abstract OpenFog RA applied to various use
cases. This will further clarify each aspect of the OpenFog RA and what
needs to be done for a successful implementation.
 Chapter 8 contains some of the open areas of fog computing and new
opportunities for research. OpenFog member companies and academic
organizations are continually enhancing and refining the OpenFog RA.
This section will help advance the overall architecture over time.
The Internet of Things, Cloud and the OpenFog RA
The Internet of Things (IoT) is driving business transformation by connecting

everyday objects and devices to one another and to cloud-hosted services.
10
Current deployment models emphasize mandatory cloud connectivity;
however, this is not feasible in many real-world situations. These are two of
the primary issues with connecting edge devices to the cloud for all services:
 Connected devices are creating data at an exponentially growing rate,

which will drive performance and network congestion challenges at the
edge of infrastructure.
 There are performance, security, bandwidth, reliability, and many
other concerns that make cloud-only solutions impractical for many
use cases.
Unfettered cloud-only architectural approaches cannot sustain the projected

data velocity and volume requirements of the IoT. To sustain IoT
momentum, the OpenFog Consortium is defining an architecture to address
infrastructure and connectivity challenges by emphasizing information
processing and intelligence at the logical edge. This approach is called fog
computing.
Figure 1 Unfettered Cloud Computing
While the cloud itself may play a vital role in many deployments, fog
computing represents a shift from traditional closed systems and a reliance
on cloud-only focused models. Fog computing is complementary to, and an
extension of, traditional cloud-based models.
The fog computing model moves computation from the cloud closer the
edge, and potentially right up to the IoT sensors and actuators. The
computational, networking, storage and acceleration elements of this new
model are known as fog nodes. These are not completely fixed to the
physical edge, but should be seen as fluid system of connectivity.
11
OpenFog and Other Consortia
The OpenFog Consortium invites open participation from across industry,

academia and non-profit organizations that have an interest in the emerging
IoT landscape. The mission of the OpenFog Consortium is complementary to
other IoT and technology industry alliance groups including the Industrial
Internet Consortium (IIC), ETSI-MEC (Mobile Edge Computing), OPC-UA,
Open Connectivity Foundation (OCF), OpenNFV, and many others. To avoid
duplication of effort and market confusion, the OpenFog Consortium will de-
emphasize efforts to optimize for some of these application spaces, and
instead focus on optimally serving vertical markets not addressed by other
initiatives. In the longer term, through liaisons with these and other bodies,
we will drive more convergence across the IoT industry under a common
view of edge and fog architectures.
Figure 2 OpenFog Consortium and Other Consortia
12
3 Use Cases for Fog
Fog computing targets cross-cutting concerns like the control of
performance, latency and efficiency are also key to the success of fog
networks. Cloud and fog computing are on a mutually beneficial, inter-
dependent continuum.
Certain functions are naturally more advantageous to carry out in fog nodes,
while others are better suited to cloud. The traditional backend cloud will
continue to remain an important part of computing systems as fog
computing emerges. The segmentation of what tasks go to fog and what
goes to the backend cloud are application specific. This segmentation could
be planned, but also change dynamically if the network state changes in
areas like processor loads, link bandwidths, storage capacities, fault events,
security threats, cost targets, etc.
The OpenFog RA enables fog-cloud and fog-fog interfaces. OpenFog

architectures offer several unique advantages over other approaches, which
we term SCALE:
 Security: Additional security to ensure safe, trusted transactions

 Cognition: awareness of client-centric objectives to enable autonomy
 Agility: rapid innovation and affordable scaling under a common
infrastructure
 Latency: real-time processing and cyber-physical system control
 Efficiency: dynamic pooling of local unused resources from
participating end-user devices
A quick use case example to illustrate the value of fog: Consider an oil
pipeline with pressure and flow sensors and control valves. One could
transport all those sensor readings to the cloud (perhaps using expensive
satellite links) analyze the readings in cloud servers to detect abnormal
conditions, and send commands back to adjust the positon of the valves.
There are several problems with this scenario: the bandwidth to transport
the sensor and actuator data to and from the cloud could cost many
thousands of dollars per month; those connections could be susceptible to
hackers; it may take several hundred milliseconds to react to an abnormal
sensor reading (during which time a major leak could spill significant oil);
and if the connection to the cloud is down, or the cloud is overloaded,
control is lost.
13
Now, consider placing a hierarchy of local fog nodes near the pipeline. They
can connect to sensors and actuators with inexpensive local networking
facilities. Fog nodes can be highly secure, lessening the hacker threat. Fog
nodes can react to abnormal conditions in milliseconds, quickly closing
valves to greatly reduce the severity of spills. Local control in the fog nodes
produces a more robust control system. Moving most of the decision-making
functions of this control system to the fog, and only contacting the cloud
occasionally to report status or receive commands, creates a superior control
system.
This document describes a set of high-level attributes of fog computing that

we call the pillars (including some of the fog advantages described in the
pipeline control scenario). These include security, scalability, openness,
autonomy, reliability, agility, hierarchical organization and programmability.
We will discuss all of these pillars in detail later in this document.
Platform as a service (PaaS) is a category of cloud computing services that

provides a platform allowing customers to develop, run, and manage web
applications without the complexity of building and maintaining the
infrastructure typically associated with developing and launching an
application. OpenFog RA defines the required infrastructure to enable
building Fog as a Service (FaaS) to address certain classes of business
challenges. FaaS includes Infrastructure as a Service (IaaS), Platform as a
Service (PaaS), Software as a Service (SaaS), and many service constructs
specific to fog. The infrastructure and architecture building blocks below
show how FaaS may be enabled and will be expanded upon in the reference
architecture.
The OpenFog RA describes a generic fog platform that is designed to be

applicable to any vertical market or application. This architecture is
applicable across many different markets including, but not limited to,
transportation, agriculture, smart-cities, smart–buildings, healthcare,
hospitality, financial services, and more, providing business value for IoT
applications that require real-time decision making, low latency, improved
security, and are network-constrained. In this section, we look at a few
specific use cases.
Transportation Scenario: Smart Cars and Traffic

Control
In 2016, the average person creates around 650MB of data every day and
by 2020, some project that to more than double. However smart
autonomous cars will generate multiple terabytes of data every day from the
combinations of light detection and ranging (LIDAR), global positioning
14
systems (GPS), cameras, etc. When the smart car is coupled with intelligent
infrastructure, it is clear that a cloud-only model will not work for
autonomous transportation, and that a fog computing approach is required.
Many of the architecture requirements we describe in smart cars and traffic
control also apply to other transportation areas, such as ships/boats, trains,
trucks, busses and drones. In this section we will highlight the opportunity
for fog computing for smart cars and traffic control, and explain how the
requirements are addressed by the OpenFog RA. The figure below is an
overview of an intelligent highway application of the OpenFog RA.
Figure 3 OpenFog Transportation: Smart Car and Traffic Control System
The smart car and traffic control use case presents an opportunity to
examine a fog environment containing a rich set of interactions among
multiple fog domains as well as multiple cloud domains. Among other things,
this use case demonstrates:
15
 A rich set of interactions among multiple fog domains, as well as
multiple cloud domains, including Element Management Systems
(EMS), service provider (SP), metro traffic services, and system
manufacturer clouds.
 Mobile fog nodes supporting vehicle-to-vehicle (V2V), vehicle-to-
infrastructure (V2I), and vehicle-to-x (V2X) interactions.
 Multiple fog networks owned and operated by different authorities
providing similar (and different) functionality
 Multi-tenancy across fog nodes will also be a burgeoning opportunity
to consolidate multiple fog networks, improving efficiency.
 Both private and public fog and cloud networks used by a single end
point device.
This use case shows also shows the hierarchical and distributed advantages
of a fog architecture. As shown in the figure above, the system includes
several types of sensors (and actuators) that we refer to as “things.”
Things include roadside sensors (infrastructure) and on-vehicle sensors.

These sensors provide data so that the various systems (lights, cars, etc.)
can carry out their given functions (e.g. vehicle driving autonomously).
Smart transportation systems also manage the actuators that control parts
of the infrastructure, such as traffic signals, gates, and digital signs.
The vehicles connect to the cloud and a hierarchy of fog nodes that service
the autonomous vehicle or traffic control systems.
Fog computing nodes in the vehicle

In this use case example, the vehicle is a mobile fog node that
communicates with other fog nodes as they become available, an example of
V2I interactions. The mobile fog node must also be capable of performing all
required in-vehicle operations autonomously in circumstances where it
cannot connect to other fog nodes or the cloud.
In-vehicle fog nodes provide services including infotainment, advanced

driver assistance systems (ADAS), autonomous driving, collision avoidance,
navigation, etc. Several different networking technologies, including
Dedicated Short Range Communications (DSRC), cellular (e.g. 3G, LTE, 5G,
etc.) and other networking technologies securely connect the vehicles to
each other and the infrastructure.
The Transportation Fog Network

The Transportation Fog Network is comprised of a three-level hierarchy of
fog nodes.
16
The first level of the hierarchy is the infrastructure fog nodes, or roadside
fog nodes. At this level, the roadside fog sensors collect data from other
devices such as roadside cameras. The fog nodes perform some local
analysis for local action, such as alerting the vehicle about poor road
conditions, triggering autonomous response to slow down, and perform
some autonomous functions, even if connections to higher layers are
unavailable. Data from the first level of interactions is aggregated and sent
up the fog hierarchy to the second and third levels of the hierarchy—
neighborhood and regional fog nodes—for further analysis and distribution.
Some of the data may also be distributed east-west to other infrastructure
nodes for their use. Typically, each fog layer in the hierarchy will provide
additional processing, storage, and network capabilities in service of the
vertical application at their level of the hierarchy. For example, higher level
layers provide additional processing to provide data analytics or large
storage capacities.
Traffic control systems

Traffic control fog nodes may receive input from other sources, such as
smart traffic light systems, municipal managers, and cloud-based systems.
Data flows between the traffic control system, infrastructure fog nodes and
vehicles in all directions, insuring all levels of the hierarchy have the data
and control capabilities they need.
The goal of the OpenFog RA for smart cars and traffic control is to ensure an
open, secure, distributed, and scalable architecture that optimizes real time
capabilities within a multi-supplier ecosystem. The transportation example
shows a complex system of autonomous things and infrastructure generating
massive amounts of data. We believe that this use case highlights the need
for fog computing to enable safe and effective operations in IoT, 5G, AI and
other advanced scenarios.
Visual Security and Surveillance Scenario
Surveillance and security cameras are being deployed worldwide. These

cameras are used to ensure security of materials, people, and places. In
addition, these cameras have the ability to generate a massive amount of
data, which can exceed terabytes per day for a single camera.
Traditional cloud models that were deployed for low-resolution cameras
aren’t scalable with the 1080p and 4K cameras because of the sheer
availability and/or cost of network transport. Additionally, decisions on
17
security need to be made at the camera or installation location and cannot
be made solely in the cloud. Machine vision is also a prime candidate for
accelerators and dynamic updating of various algorithms in both hardware
and software. These cameras are capturing images of people, places, or
things and are tightly coupled to decision–making, which requires a
heightened level of security of the camera’s software and hardware assets.
Smart cities, smart homes, retail stores, public transportation,

manufacturing and enterprises increasingly rely on camera sensors to secure
people, identify unauthorized access, and increase safety, reliability and
efficiency. The sheer bandwidth of visual (and other sensor) data being
collected over a large-scale network makes it impractical to transport all the
data to the cloud to obtain real-time insights. A particularly demanding
application space is surveillance of high value installations with many people
and objects moving through them. Controlling a large network of cameras in
an airport is a good example of such an application. This use case will be
studied in greater detail later in this document to illustrate applying the
architecture to a concrete application.
City-scale deployments that include placing cameras on traffic lights and

other camera deployments in remote areas don't have high-bandwidth
connectivity to the cloud to upload the collected video, even if the video
could fit over the network infrastructure. Real-time monitoring and detection
of anomalies (intruders into a building, the fall of an elderly citizen, the
misfiring of a piece of manufacturing equipment) pose strict low latency
requirements on surveillance systems; timeliness is important from the
standpoint of both detection and response.
Additionally, privacy concerns must be addressed when using a camera as a

sensor that collects image data so that the images do not reveal a person's
identity or reveal confidential contextual information (e.g. intellectual
property in a manufacturing plant) to any unauthorized parties. OpenFog RA
deployments provide the opportunity to build real-time, latency-sensitive
distributed surveillance systems that maintain privacy. It leverages fog
nodes to intelligently partition video processing between fog nodes co-
located with cameras and the cloud so as to enable real-time tracking,
anomaly detection, and insights from data collected over long time intervals.
Video analytics algorithms can be located on fog nodes close to the cameras,
and take advantage of the heterogeneous processor capability of fog,
running parts of the video analytics algorithm on conventional processors or
accelerators.
18
The visual security for airports use case requires all of these fog capabilities
to meet its performance, reliability, security and efficiency goals. This
includes vehicle detection, people detection, smart retail, and other areas
where machine vision via video analytics is important for fog computing.
Please see the detailed analysis in chapter 7 for many more details including
an application of the OpenFog RA.
Smart Cities Scenario
Figure 4 Opportunities for Smart Cities
Smart cities are using technology to deal with many challenges, including
traffic congestion, public safety, energy consumption, sanitation, and public
internet connectivity.
The OpenFog RA enables greater efficiency and economic realities of smart

city operations. The figure above illustrates the various aspects where fog
computing can impact smart cities including but not limited to:
 Intelligent city with smart parking, shopping, and infrastructure.

 Intelligent hospitals linking all aspects for greater patient care and
healthcare delivery.
 Intelligent highway systems to optimize utilization of infrastructure.
 Intelligent factories and software defined industrial systems.
Connectivity
19
While most modern cities have one or more cellular networks providing city-
wide coverage, these networks often have capacity and peak bandwidth
limits that just barely meet the needs of their existing subscribers. This
leaves little bandwidth for the advanced municipal services envisioned in a
smart city. OpenFog RA deployments coupled with 5G technologies provide
an opportunity to address this concern. Fog nodes can provide local
processing and storage, and optimize network usage.
Safety and Security
Smart city planning also includes critical public safety and security
requirements. For example:
 Municipal networks carry sensitive traffic and citizen data (e.g., police
dispatches), and operate life-critical systems (e.g., first responder
communications)
 Video security and surveillance systems capture suspicious or unsafe
conditions (e.g., utility network problems, unauthorized use of public
spaces, etc.)
Note that smart cars and traffic congestion, which is covered as a separate
use case, are also top priorities for smart cities.
By providing secure data and distributed analytics, fog computing will play a
key role in addressing public safety and security issues for smart cities.
3.3.1 Smart Buildings
Smart buildings may contain thousands of sensors to measure various

building operating parameters, including temperature, humidity, occupancy,
door open/close, keycard readers, parking space occupancy, security,
elevators, and air quality. These sensors capture telemetry data at various
intervals and transmit this information to a local storage server. Once this
information is processed (analyzed), controller-driven actuators will adjust
building conditions as necessary.
Some of this processing and response is extremely time-sensitive. For

example, turning on fire suppression systems in response to a fire event or
locking down an area if an unauthorized person tries to gain entry. Time-
sensitive means real-time response, which requires processing in close
proximity to the infrastructure devices.
20
The OpenFog RA model can be extended into the building’s control hierarchy
to create a number of smart, connected spaces within each building. Using
the hierarchical design of the OpenFog RA, each floor, wing, or even
individual room could contain its own fog node.
A fog node could be responsible for:
 Performing emergency monitoring and response functions.

 Performing building security functions.
 Controlling climate and lighting.
 Providing a more robust compute and storage infrastructure for
building residents to support smartphones, tablets and desktop
computers.
Locally stored operational history can be aggregated and sent to the cloud
for large-scale analytics. These analytics can be applied to machine learning
to create optimized models, which are then downloaded to the local fog
infrastructure for execution.
Additional Use Cases
Fog computing is relevant to many more use cases in addition to those

highlighted here. The OpenFog Board of Directors and Technical Committee
will set the most important areas for the Consortium and its partners to
focus on in the near and long term. We will continue to expand upon all use
cases in further discussions, publications, and testbeds.
21
4 Pillars of OpenFog RA
The OpenFog RA is driven by a set of core principles called pillars. These
pillars form the belief, approach and intent that guided the definition of the
reference architecture. They represent the key attributes that a system
needs to embody the OpenFog definition of a horizontal, system-level
architecture that provides the distribution of computing, storage, control,
and networking functions closer to the data source (users, things, et al)
along the cloud-to-thing continuum.
The following sections describe each pillar of the architecture.
Figure 5 Pillars of OpenFog
22
Security Pillar
Many IoT applications supported by the OpenFog RA have privacy critical,
mission critical, and even life critical aspects. As such, any security
compromise in the fog network can have severe consequences. The OpenFog
RA, as it abstracts these technologies, will enable the flexible creation of
computing environments that address a broad spectrum of security concerns
spanning from IoT devices to cloud and the fog networks in between.
Security in the OpenFog RA is not a one-size-fits-all architecture. Rather, it

describes all of the mechanisms that can be applied to make a fog node
secure from silicon to software application. Business case, target market,
and vertical use case, as well as the location of the node itself, will all create
a set of requirements for that node. However, there are certain foundational
parts of the architecture, which must be in place in order to build a secure
execution environment.
Security implementations have many different descriptions and attributes

such as privacy, anonymity, integrity, trust, attestation, verification, and
measurement. These are key attributes for the OpenFog RA. Achieving the
foundational elements for security requires an approach to discover, attest,
and verify all smart and connected “things” before trust can be established.
Conformance to the OpenFog RA requirements will ensure that an OpenFog

deployment will be built on a secure end-to-end compute environment. This
includes the OpenFog node security, OpenFog network security, and
OpenFog management and orchestration security. This will allow architects
and designers to focus on the high-value security and privacy problems
specific to the types of devices used in their application.
In many applications, particularly for brownfield deployments, or for tiny

devices and sensors with little to no security capability, an OpenFog node
may act as a device’s first point of secure entry into an OpenFog compute
hierarchy and the cloud.
The security pillar of the OpenFog RA starts with a clear definition of base
building blocks. All fog nodes must employ a hardware-based immutable
root of trust. The Hardware Root of Trust is a trusted hardware component
which receives control at power-on. It then extends the chain of trust to
other hardware, firmware, and software components. The root of trust
should then be attestable by software agents running within and throughout
the infrastructure. Because of the proximity to the edge, nodes in fog
networks often act as the first node of access control and encryption. This
23
means they must provide contextual integrity and isolation, and control
aggregation of privacy-sensitive data before it leaves the edge.
As more complex topologies are created in FaaS implementations, the

attestation continues as a chain of trust from the fog node, to other fog
nodes, and to the cloud. Since fog nodes may also be dynamically
instantiated or torn down, hardware and software resources should be
attestable. Components that are not attestable should not be fully allowed to
participate in the fog node or may not be deemed to have fully trustworthy
data.
Scalability Pillar
The scalability pillar addresses the dynamic technical and business needs
behind fog deployments. Elastic scalability cuts across all fog computing
applications and verticals. The hierarchical properties of fog and its location
at logical edges of networks add additional scaling opportunities.
 Individual fog nodes can scale internally, through the addition of

hardware or software.
 Fog networks can scale up and out through the addition of fog nodes
to assist heavily loaded nodes, either on the same level of the fog
hierarchy or in adjacent levels.
 A network of fog nodes can be scaled up or down in a demand-driven
elastic environment.
 Storage, network connectivity, and analytics services can scale with
the fog infrastructure.
Because of the variability in the use cases for fog computing, the OpenFog
RA enables elastic scaling of modest deployments through large mission
critical deployments based on demand. This scalability is essential for fog
computing implementations to adapt to workload, system cost, performance,
and other changing business needs.
Scalability may involve many dimensions in fog networks:
 Scalable performance enables growth of fog capabilities in response

to application performance demands (e.g., reducing latency between
sensor reading and resulting actuator responses).
 Scalable capacity allows fog networks to change in size as more
applications, endpoints, “things,” users, or objects are added or
removed from the network. You can add capacity to individual fog
nodes by adding hardware like processors, storage devices, or network
24
interfaces. You can also add capacity through software and various
pay-as-you-grow licensing. The converse is also true.
 Scalable reliability permits the inclusion of optional redundant fog
capabilities to manage faults or overloads. Redundant fog nodes also
ensure a large deployment’s integrity and reliability at scale, which is
part of the reliability, availability, and serviceability (RAS) pillar. There
are hardware and software aspects to scalable reliability. The
scalability mechanisms supporting the reliability of fog networks must
themselves be highly reliable. Availability (which is a scaling measure
closely related to reliability) scales through similar methods.
 Scalable security is often achieved through the addition of modules
(hardware and software) to a basic fog node as its security needs
become more stringent. Capabilities like scalable distribution, rights
access, crypto processing capacity, and autonomous security features
contribute to scalable security.
 Scalable hardware involves the ability to modify the configuration of
the internal elements of fog nodes, as well as the numbers of and
relationships between fog nodes in networks.
o Processors scale from modest single core CPUs to specialized
accelerator chips with thousands of cores or millions of gates.
o Networking scales from a single wireless or wire line interface to
large arrays of wireless, wire line, and fiber interfaces with
aggregate capacities of many Gb/s.
o Storage can scale from a simple flash chip to large arrays of
flash / rotating disks and network attached file systems.
These resources can be configured in initial deployments and retrofit into

existing modular fog nodes on an as-needed basis. It is also possible to scale
at the network level, by adding arrays of fog nodes at locations in the
network where single nodes formally managed the entire load. Hardware
scaling can also be in the downward direction, where modules or entire fog
nodes that are no longer are needed at a specific location are powered down
and/or removed (and perhaps reused elsewhere in a fog network where
there is a higher need.
 Scalable software is also important and includes applications,

infrastructure, and management.
o The management infrastructure of fog must scale to enable the
efficient deployment and ongoing operation of tens of millions of
fog nodes in support of billions of smart and connected things.
o Orchestration must be scalable to manage the partitioning,
balance, and allocation of resources across the fog network.
o Analytics as a capability of fog networks has particularly
aggressive scalability targets. This is because analytics
25
algorithms undergo several orders of magnitude scaling due to
increased capacity demands and several additional orders of
magnitude due to ever-increasing sophistication of the analytics
algorithms.
o Composability and modularity are important aspects of
scalability, where individual hardware and software components
(perhaps numbering in thousands of types and millions of
instantiations) are assembled into a fog network optimized to
run the specific applications required.
Scalability enables fog nodes to provide basic support to address the

business requirements and enable a pay-as-you-grow model for the FaaS,
which is essential to the economics of its initial deployment and long-term
success.
Openness Pillar
Openness is essential for the success of a ubiquitous fog computing

ecosystem for IoT platforms and applications. Proprietary or single vendor
solutions can result in limited supplier diversity, which can have a negative
impact on system cost, quality and innovation. The openness pillar
importance is highlighted in our desire for fully interoperable systems,
supported by a vibrant supplier ecosystem.
Openness as a foundational principle enables fog nodes to exist anywhere in

a network and span networks. This openness enables pooling by discovery,
which means that new software-defined fog nodes can be dynamically
created to solve a business mission. The security pillar shares a common
theme and requirements in openness characteristics
 Composability supports portability and fluidity of apps and services

at instantiation. Additional emphasis of composability is visible in the
programmability pillar.
 Interoperability leads to secure discovery of compute, network, and
storage and enables fluidity and portability during execution. The
marketplace has clearly articulated its desire for a vibrant supplier
ecosystem, with reasonable expectations that elements from one
supplier can be freely substituted for elements from another supplier.
This will be addressed through testbeds, FogFests (plug fest),
standardization, and open implementations.
 Open communication enables features like pooling of resources near
the edge of the network to collect idle processing power, storage
capacity, sensing ability, and wireless connectivity. For example, a
26
compute-intensive application developed in fog architecture can
leverage hundreds of gigabytes sitting idle on nearby laptops,
systems, and set-top boxes in a household every evening, or among
the passengers of a public transit system. The open discovery of these
nearby compute resources is critical. Doing the functional work nearest
the edge avoids additional network taxes when moving up the stack
towards the cloud. We define network taxes as the cost of
transmission.
 Location transparency of any node instance to ensure that nodes
can be deployed anywhere in the hierarchy. Location transparency
provides an alternative to network operator control. This means that
any IoT device, such as a smart watch, does need its own carrier-
owned data plan. Each thing or software entity can observe its local
conditions and make decisions on which network to join. Each endpoint
in a fog network can optimize its path to the computational,
networking and storage resources it needs (no matter if those
resources are in the hierarchical layers of the fog, or in the cloud).
Autonomy Pillar
The autonomy pillar enables fog nodes to continue to deliver the designed
functionality in the face of the external service failures. In this architecture,
autonomy is supported throughout the hierarchy. Decision making will be
made at all levels of a deployment’s hierarchy including near the device or
higher order layers. Centralized decision-making in the cloud is no longer the
only option. Autonomy at the network edge means intelligence from local
devices and peer data can be used to fulfill the business’ mission at the point
where it makes the most business sense.
The OpenFog RA supports autonomy for a wide range of functions. It does

not rely upon centralized entity for operation (e.g., a backend cloud). Some
of the typical areas for autonomy at the edge include:
 Autonomy of discovery to enable resource discovery and

registration. For example, an IoT device coming online in the field
would typically “phone home” first to let the backend cloud know it is
alive and its associated functions are available. But when an uplink
network to the cloud is unavailable, it can stop the device from going
live. An autonomous fog node can potentially act as a proxy for the
device registration, which then allows the device to come online
without the backend cloud.
 Autonomy of orchestration and management (O&M) automates
the process of bringing services online and managing them through
27
the operational lifecycle and decommissioning. Autonomy of O&M
entails a number of actions including: instantiation of services;
provisioning the environment around the services, such as routing of
data flows; and keeping track of the health and status of the
resources. All these actions should be as automated as possible
through programmability and policies. The architecture includes an
autonomous and scalable O&M function that is set up to handle any
surge of demand for resources, without real-time reliance on the cloud
or the need for significant human labor.
 Autonomy of security enables devices and services to come online,
authenticate themselves against a minimal set of fog security services,
and perform their designed functions. In addition, these security
services can store records for future audits. With autonomy, these
actions can be performed where they are needed, when they are
needed, and regardless of connectivity to the cloud. Fog nodes can
autonomously react to evolving security threats, such as updating
virus screening algorithms, determination of denial-of-service (DoS)
attacks, etc. without administrator involvement.
 Autonomy of operation supports localized decision making by IoT
systems. Sensors provide data, which is the basis for autonomous
actions at the edge. If the cloud or a single place in the system’s
hierarchy is the only location where decisions can be made, this
violates the ability to ensure reliability and as such, the architecture
ensure operational autonomy.
 Cost savings is a key motivator for autonomy. Connectivity today
costs money. The more data that is sent through the network, the
higher the costs are for businesses due to network taxes. This drives
the need for more processing at the edge of the network, with just-in-
need and just-in-time data sent to the cloud as required for additional
business insights. For example, when an oil rig generates 30,000 data
points a second, not all of the data must be sent through an expensive
satellite link. Local and fog domain analytics and pre-processing can
autonomously filter out the unimportant data points and extract the
more mission critical ones to be delivered to the next layer in the
hierarchy.
A key aspect of fog computing is turning data into actionable wisdom. We

term this DIKW, which stands for “Data gathered becomes Information
when stored and retrievable becomes Knowledge. Knowledge enables
Wisdom for autonomous IoT.” This principle is the basis for localized
analytics to enable autonomous decision making nearest the edge.
28
Programmability Pillar
The programmability pillar enables highly adaptive deployments including

support for programming at the software and hardware layers. This means
that re-tasking a fog node or cluster of fog nodes for accommodating
operational dynamics, can be completely automated. The re-tasking can be
done with the help of the fog nodes inherent programmability interfaces
which we describe using general purpose compute or accelerator interfaces.
Programmability of a fog node includes the following benefits:
 Adaptive infrastructure for diverse IoT deployment scenarios and

support changing business needs.
 Resource efficient deployments maximizing the resources by using
a multitude of features including containerization. This increases the
portability of components and is a key design goal enabled by
programmability.
 Multi-tenancy to accommodate multiple tenants in a logically isolated
runtime environment.
 Economical operations that results adaptive infrastructure to
changing requirements.
 Enhanced security to automatically apply patches and respond more
quickly to evolving threats.
Reliability, Availability, and Serviceability (RAS)

Pillar
Reliability, availability, and serviceability (RAS) is resident throughout

successful system architectures and, as such, takes on great importance in
the OpenFog RA. Hardware, software, and operations are the three main
areas of the RAS pillar.
A reliable deployment will continue to deliver designed functionality under

normal and adverse operating conditions. The reliability of the RAS pillar
includes but is not limited to the following properties:
 Ensuring reliable operation of the underlying hardware upon which the

software is operating, enabling reliable and resilient software and a
reliable fog network, which is generally measured in uptime.
 Safeguarding the availability and integrity of data and compute on
edge gateways using enhanced hardware, software, and network
designs.
 Autonomous predictive and adaptive self-managing capabilities when
required by the health of the system to initiate self-healing routines for
29
hardware and software and upgrade new firmware/application and
security patches.
 Increasing customer satisfaction by simplifying support and device
self-optimization and healing.
 Initiating requests for preventative maintenance, including new
hardware and software patches, network re-routing, etc.
 Testing and validation of system components, including device drivers
and diagnostic tools under a variety of environmental conditions.
 Providing alarms, reports, logs, etc.
 Validation of system platforms and architectures through
interoperability certification test suites.
Availability ensures continuous management and orchestration, which is

usually measured in uptime. The availability of the RAS pillar includes but
not limited by the following properties:
 Secure access at all levels of a fog hierarchy for orchestration,
manageability, and control, which includes upgradeability, diagnostics
and secure firmware modification.
 Fault isolation, fault syndrome detection, and machine learning to help
improve Mean Time To Repair (MTTR) of a failed system to achieve
higher availability.
 Concept of cloud based back-end support with availability of interfaces
throughout the system.
o Secure remote access from a plurality of devices (not just a
single console).
o Redundant/duplicate device (peer-to-peer) IoT platform.
o Mesh access capabilities of end-point sensor/peering.
o Remote boot capabilities of the platform.
 Modification and control from the lowest level firmware
(BIOS) through to the highest software in the hierarchy
(cloud).
o Support for redundant configurations for persistent productivity.
Servicing a fog deployment ensures correct operation. Serviceability of the

RAS pillar includes but is not limited by the following properties:
 Highly automated installation, upgrade, and repair to efficiently deploy

fog computing at scale.
 Hardware or software can either autonomously heal or be serviced by
the various manufacturers.
 Ease of use to accommodate maintenance.
 Serviceability of the system:
o Hardware, software, applications, networking, and data
30
o Ease of access/swap-out of the hardware (component
interoperability).
o Ease of secure upgradeability of software, BIOS, and applications
locally or remotely and in real time.
o Replication of system configuration over cloud on replaced/swap-
out systems.
 Support for redundant configurations for persistent productivity.
RAS is especially important for OpenFog RA deployments in harsh

environmental conditions and remote locations. This is why aspects from
RAS are found throughout the architecture.
Agility Pillar
The agility pillar addresses business operational decisions for an OpenFog RA

deployment. It is not possible for humans alone to analyze the data
generated at the scale predicted by IoT as the basis for rapid, sound
business and operational decisions. The agility pillar focuses on transforming
this volume of data into actionable insights. Agility also deals with the highly
dynamic nature of fog deployments and the need to respond quickly to
change.
Data generation by sensors and systems in an OpenFog RA deployment are

turbulent, bursty, and are often created in huge volumes. Most importantly,
data may not have context, which is created only when the data is collated,
aggregated, and analyzed. The analysis of data can be executed at the cloud
level, but this subjects the data to increasing levels of latency. The ideal
approach is to make operational decisions as soon as data can be turned into
a meaningful context. The architecture enables the creation of context close
to the data generation where it makes the most sense for a given scenario.
More strategic, system-wide decisions and policy management can be made
further up the layers in the fog hierarchy. This avoids network dependencies
we termed as “network taxes” as described in other OpenFog RA pillars.
OpenFog architectural approaches allow IoT system developers to optimize

the placement of their applications as decision making components.
Hierarchy Pillar
Computational and system hierarchy is not required for all OpenFog

architectures but it is still expressed in most deployments. The OpenFog
architecture is complementary to traditional cloud architectures due in part
to the OpenFog hierarchy pillar.
31
OpenFog RA computing resources can be seen as a logical hierarchy based
on the functional requirements of an end-to-end IoT system. Depending on
the scale and nature of the scenario being addressed, the hierarchy may be
a network of smart and connected partitioned systems arranged in physical
or logical layers, or it may collapse into a single physical system (scalability
pillar).
Using building automation from smart cities as an example, a company that

manages a single office complex may have the entire fog deployment
located locally. A large commercial property management company may
have distributed fog deployments at local and regional levels feeding
information to centralized systems and services. Each fog node is
autonomous (autonomy pillar) to ensure uninterrupted operations of the
facility it manages.
The figure below shows a logical view of the IoT system from a
computational perspective. Each layer in the hierarchy addresses a specific
concern of the IoT system.
Figure 6 Layered Architecture View of an IoT System
Devices in the hierarchy: Sensors and actuator devices are the physical
things and produce telemetry data to be consumed by the monitoring and
control layer. This layer analyzes the telemetry and generates actuation
commands if the process being monitored deviates from the desired state.
Note that the term “process” is used in an abstract sense that it is
represented by a set of measured parameters that depend on a set of
actuator settings. Depending on the domain problem, some systems may
not have any actuators. Similarly, for scenarios like mobile network
acceleration, the core function is to accelerate content delivery and not
32
monitoring and control. On the other hand, systems like building operations
may have actuators to change the HVAC and lighting based on occupancy.
Monitoring and control in the hierarchy: Sensors and actuators are

connected to microcontrollers that are programmed to monitor and control
the state of the processes. A process state is represented by a set of
parameters measured by the sensors and modified by the actuators. The
main responsibility of this layer is to execute control logic through stateful
inspection of the sensor telemetry. This involves computing alarms and
generating events, which may trigger workflows through machine-to-
machine or human intervention.
Operational support in the hierarchy: The operational support layer is

responsible for analyzing streaming telemetry and storing operationally
oriented analytics. The analytics may be presented through interfaces like
control room dashboards and mobile applications. The scope of the analytics
at this layer is narrow; it focuses on the operational aspects of the physical
environment for which the system is responsible. This layer combines drill
down historical analytics with streaming analytics for a composite picture of
real-time operations with some short-term history. The agility pillar is seen
in the hierarchy as the implementation of complex event processing on the
streaming telemetry data.
Surrogacy in the hierarchy: The computation of a complex operation in

the fog nodes may be delegated to the hierarchical nodes to leverage
adjacent resources. Consider virtual reality tasks associated with a wearable
such as smart glasses. Some of the information retrieval and computation
tasks may be carried out on the glasses, while an associated element in the
hierarchy (e.g., a smartphone) may handle its storage and connectivity
requirements. This hierarchical architecture may leverage all of these
devices at the same time, with an intelligent division of labor across them.
Business support in the hierarchy: The primary responsibility of this

layer is to store and analyze the entire history of the IoT operations that
span multiple systems. This is the system of record for IoT operations as
governed by the compliance and record retention policies. Petabyte scale
analytics will help in mining insights, business planning, comparing the
operational efficiency of processes, operational optimization through training
machine learning models, etc. Additionally, metadata and reference data
management, business rule management, and the operational health of
lower layers are the other aspects of this layer. These are also viewed in the
agility pillar.
33
4.8.1 Hierarchical Fog Deployment Models
The figures below show a subset of the combination of fog and cloud
deployed to address various domain scenarios as framed by the layered view
of IoT systems. Each fog element may represent a hierarchy of fog clusters
fulfilling the same functional responsibilities. Depending on the scenario,
multiple fog and cloud elements may collapse into a single physical
deployment. Each fog element may also represent a mesh of peer fog nodes
in use cases like connected cars, electrical vehicle charging, and closed loop
traffic systems. In these use cases, fog nodes may securely discover and
communicate with each other for exchanging context-specific intelligence.
Figure 7 IoT System Deployment Models
Figure 1 shows a fog deployment hierarchy that is independent of the cloud.

This model may be applicable for use cases where cloud can’t be used for
34
reasons such as low event to action time window, regulatory compliance,
military grade security and privacy, and unavailability of a central cloud in a
particular geography. Examples include armed forces combat systems, drone
operations, some healthcare systems, hospitals, and ATM banking systems.
Figure 2 shows the cloud used for information processing related to decision
making that may have event-to-action time window ranging from hours to
days to months. Operation-centric information processing is done by fog
deployments located close to the infrastructure/process being managed. Use
cases include commercial building management, commercial solar panel
monitoring, and retail.
Figure 3 shows the local fog infrastructure used for time-sensitive

computation, while the cloud is used for the balance of operational and
business-related information processing. Use cases include commercial UPS
device monitoring, mobile network acceleration, and content delivery
networks (CDNs) for Internet acceleration.
Figure 4 supports use cases like agriculture, connected cars, and remote
weather stations. These use cases leverage the cloud for the entire stack
due to the constrained environments in which the deployment of fog
infrastructure may not be feasible or economical. Fog nodes at the device
layer may get some of the monitoring and control function for safety related
control. The enterprise systems integrate with cloud for business operations.
Note that the functional boundaries shown in Figures 1-3 are fluid and can
be physically deployed in multitude of combinations based on the
architecture of the domain-specific solutions. In real world deployments as
we discussed earlier, there may be many combinations of physical
deployments that involve multi-tenants, fog, and cloud deployments owned
by multiple entities. The following diagram illustrates some those models:
35
Figure 8 Fog Hierarchy Example
Figure 9 Fog Hierarchical Deployment Model
36
Many of the usages will occur as represented in Figures 2 and 3. The three-
layer fog hierarchies shown here are for illustrative purposes only. Real-
world fog deployments may have more or fewer levels. Different vertical
application use cases may use a fog hierarchy differently. For example, in a
smart city, there may be fog nodes in a region, neighborhood, street corner,
and building level. In a smart factory, the hierarchy may be divided by
assembly lines, manufacturing cells, and machines.
37
5 Reference Architecture Overview
The OpenFog reference architecture is based on the eight pillars described in

the previous chapter. We use the ISO/IEC/IEEE 42010:2011 international
standard as the guideline for describing architecture to stakeholders. The
standard enables a common vocabulary across IoT to help support cross-
organizational technical collaboration. These are some of the shared
vocabulary terms used in the OpenFog RA:
 OpenFog Architecture Description: is an abstract representation of

an instance of a fog node. It is a composite of multiple views we used
to address stakeholders in the fog computing value chain.
 Viewpoint: A viewpoint is a way of looking at a system. These
included but are not limited to Functional and Deployment viewpoints.
 View: A view is a representation of one or more structural aspects of
the architecture. In the current revision of the OpenFog RA, the
structural aspects are the Software view, System view, and Node view.
 Perspective: A perspective is a cross-cutting concern of the
architecture.
Functional Viewpoint
The functional viewpoint of the architecture shows how we apply the

OpenFog architectural elements and views to address the various concerns
of the stakeholders to satisfy a given scenario. Each scenario chosen will
focus on a different aspects and market opportunities for fog computing. We
fully expect that the architectural description, views, and perspective may
change over time. This change and refinement should be driven from
testbeds and application of the OpenFog RA to multiple different markets we
see as important to fog computing.
The first end-to-end scenario we will address is the visual security scenario.
Our intent will be to define how that scenario works, and then work through
various testbeds associated with those aspects to validate or modify the
architecture.
38
Deployment Viewpoint
5.2.1 OpenFog Deployment Types
How fog software and fog systems are deployed to address a given scenario
is also very important. There are many deployment types ranging from
embedded to large clustered systems that are fully interconnected. The
deployment type(s) chosen are scenario specific, but key aspects of the
architecture remain visible regardless of deployment type. However, some
may grow or shrink in importance.
5.2.2 N-Tier Fog Deployment
In most fog deployments, there are usually several tiers (N-tiers) of nodes.
For this example, we will use a simple food processing plant to help solidify
the logical tiers. There is a conveyor belt on which food is processed before
moving on to the next level of packaging and shipment.
Figure 10 Multi-Tier Deployment
 Nodes at the edge are typically focused on sensor data

acquisition/collection, data normalization, and command/control of
sensors and actuators.
o Using our example, the fog nodes nearest the physical
technology edge (at the conveyor belt) will need to operate at
39
millisecond and sub-millisecond granularity from sensing to
actuation to avoid product contamination and to ensure safety of
operation.
 Nodes in the next higher tier are focused on data filtering,
compression, and transformation. They may also provide some edge
analytics required for critical real time or near real time processing. As
we move away from the true network edge, we see higher level
machine and system learning (analytics) capabilities.
o Using our previous example, we can easily see that this tier
needs to operate on a slightly higher level. It should be ensuring
that the conveyor belt and others around it are operating more
efficiently. This may still be in the latency of milliseconds. The
key aspect of OpenFog RA to enable migration of applications,
and functionality between the lowest tiers and the middle tiers
as required by the scenario.
 Nodes at the higher tiers or nearest the backend cloud are typically
focused on aggregating data and turning the data into knowledge. It’s
important to note that the farther from the true edge, the greater the
insights that can be realized.
Note: In some deployment models, some degree of analytics may be located

in nodes at the edge of the network (e.g. video analytics on surveillance
cameras). This is because the network pipes may not be large enough to
cost effectively carry the raw sensor data to higher layer fog nodes for
processing. In reality as computational capabilities grow, the analytics
functions at the lower tiers will grow. This will enable the overall growth of
intelligence of fog deployments over time.
Machine Learning is in the forefront of research today that requires

computation for both training models, and inference or scoring those models
for close to real time response at the edge. We could use machine learning
to optimize operations at a train station in a smart city. In the train station,
we could monitor and sense occupancy, movement, and overall system
usage and over time adapt our infrastructure to determine how to most
efficiently use it.
Using the smart city as a continued example, by including more buildings

and having them talk to each other and using another tier above them to
gain additional insights, city blocks can operate more efficiently. Learnings
from blocks of buildings provide insights on how to make the overall city
more efficient and so on and so forth. The key message is that as you move
farther away from the true network edge, you can gain operational insights
and increase the overall system intelligence. Additionally, migration of data
40
between these fog nodes in both the horizontal and vertical increases the
system performance and operational capabilities.
5.2.2.1 How Use Cases Determine the Number of Tiers
Fog deployments will come large-scale and small, based upon the given
scenario being addressed. The number of tiers in a fog deployment will be
dictated by the scenario requirements, including:
 Amount and type of work required by each tier

 Number of sensors
 Capabilities of the nodes at each tier
 Latency between nodes and latency between sensors and actuation
 Reliability/availability of nodes
In general, each level of the N-tier environment would be sifting and

extracting meaningful data to create more intelligence at each level. Tiers
are created in order to deal efficiently with the amount of data that needs to
be processed and provide better operational and system intelligence. At the
highest level this can be represented by the figure below.
Figure 11 Intelligence from data
Important aspects we will address overtime is that the software running at

every level and each node should be enabled to migrate across nodes, span
41
physical instantiations of hardware nodes, and change over time to address
the needs of a given scenario. For this to be securely and safely achieved we
need to address not only the software, but also the hardware on which that
software is executed.
5.2.2.2 Fog Node Uniformity
The architectural elements of a node will vary based on its role and position
within an N-tier fog deployment. As described previously, nodes at the edge
may be architected with less processing, communications, and storage than
nodes at higher levels. However, I/O accelerators required to facilitate
sensor data intake at the edge may be much larger in aggregate than I/O
accelerators designed for higher level nodes.
Fog nodes may be linked to form a mesh to provide load balancing,

resilience, fault tolerance, data sharing, and minimization of cloud
communication. Architecturally, this requires that fog nodes have the ability
to communicate laterally (peer to peer or east to west) as well as up and
down (north to south) within the fog hierarchy. The node must also be able
to discover, trust, and utilize the services of another node in order to sustain
RAS. Using our building and city example as before this is represented by
the following figure.
42
Figure 12 Fog Node East/West Communication
Each building is connected, neighborhoods, and regions are connected to

provide an infrastructure that may be optimized for service delivery.
OpenFog Architecture Description
As we previously described, fog computing is critical because it enables low

latency, reliable operation, and removes the requirement for persistent cloud
connectivity to address many of today’s emerging scenarios. We also
described how fog nodes can be connected partially or fully to enhance the
overall system intelligence and operation, and how system wide intelligence
grows the farther away from raw data processing.
The next step is to describe the requirements for each stakeholder in the fog
computing continuum. This includes the silicon manufacturer, system
manufacturer, system integrator, software manufacturer, and application
developer. We also believe that this architecture will help align the various
disparate edge based computing but potentially divergent work under a
singular vernacular so that we can have a common baseline and work
towards fulfilling our desire of a multi-vendor interoperable fog computing
ecosystem. The OpenFog RA description is a composite representation of
these various stakeholder concerns which we call views. We have primarily
identified these stakeholders and their associated views because they are
required to facilitate any successful fog based deployment. Before going into
43
the lower level details of the view we believe it is important to first look at
the composite architecture description.
Figure 13 Architecture Description with Perspectives
The abstract architecture includes perspectives, shown in grey vertical bars

on the sides of the architectural description. The perspectives include:
 Performance: Low latency is one of the driving reasons to adopt fog

architectures. There are multiple requirements and design
considerations across multiple stakeholders to ensure this is satisfied.
This includes time critical computing, time sensitive networking,
network time protocols, etc. It is a cross cutting concern because it
has system and deployment scenario impacts.
 Security: End-to-end security is critical to the success of all fog
computing deployment scenarios. If the underlying silicon is secure,
but the upper layer software has security issues (and vice versa) the
solution is not secure. Data integrity is a special aspect of security for
devices that currently lack adequate security. This includes intentional
and unintentional corruption.
 Manageability: Managing all aspects of fog deployments, which
include RAS, DevOps, etc., is a critical aspect across all layers of a fog
computing hierarchy.
 Data Analytics and Control: The ability for fog nodes to be
autonomous requires localized data analytics coupled with control. The
actuation/control needs to occur at the correct tier or location in the
hierarchy as dictated by the given scenario. It is not always at the
physical edge, but may be at a higher tier.
44
 IT Business and Cross Fog Applications: In a multi-vendor
ecosystem applications need the ability to migrate and properly
operate at any level of a fog deployment’s hierarchy. Applications
should also have the ability to span all levels of a deployment to
maximize their value.
As previously discussed, the OpenFog RA description is a composite of

perspectives and multiple stakeholder views used to satisfy a given fog
computing deployment or scenario. The three views that we have identified
include Software, System, and Node.
 Software view: is represented in the top three layers shown in the

architecture description, and include Application Services, Application
Support, and Node Management (IB) and Software Backplane.
 System view: is represented in the middle layers shown in the

architecture description, which include Hardware Virtualization down
through the Hardware Platform Infrastructure.
 Node view: is represented in the bottom two layers shown in Figure

19, which includes the Protocol Abstraction Layer and Sensors,
Actuators, and Control.
Note: The fog platform coupled with the fog software creates the complete
fog node. One or more fog nodes comprises a solution in a given market
segment or scenario.
However, high-level architectures, including the OpenFog RA, are intended

to help engineers, architects, and business leaders understand their specific
requirements and how fog nodes can be applied to a given scenario. The
goal of the OpenFog Consortium is to increase the number of market
segments (use cases) for fog computing, and its business value. OpenFog
will create test-beds to adapt the high-level architecture to these market
segments. These testbeds will also provide opportunities for FogFests (plug
fests) to help drive component level interoperability and accelerate time to
market.
The following sections go into more detail about the structural aspects
(views) and perspectives of the RA.
45
Perspectives (Cross Cutting Concerns)
Cross-cutting perspectives are employed throughout fog implementations.

“Cross cutting” refers to capabilities that cut across architectural layers. The
figure below shows the five cross-cutting perspectives of the fog computing
architecture.
Figure 14 OpenFog Architecture Perspectives
5.4.1 Performance and Scale Perspective

When fog computing brings some of the intelligence of cloud-based
applications and analytics to the edge of the network (or as close as possible
to the data source), the performance of the overall system (however that
system is defined) will improve. Fog computing will also enable the system
to better adapt to changing traffic patterns. This means that performance
improvements happen faster and are also more relevant and specific to
business case requirements.
Another requirement for performance is that improvements in one area must

not interrupt or slow other processes requiring a guaranteed quality of
service or performance. When measuring a fog node’s performance, we
usually look at throughput and latency. Both depend on the ability to
prioritize traffic types or classes throughout the whole system.
In the RA, virtualization and containerization technologies are both used in

fog computing to help with scalability and isolation. These newer
technologies further support the ability to assign higher priority or more
46
resources to specific applications or services dynamically. For example, high
priority network traffic can be marked and classified by the network
interface, the computational elements, and the appropriate higher-layer
applications.
Network bandwidth and local storage can also be given higher priority on a
dynamic basis. For example, if the fog node is used for traffic inspection,
CPU/memory is assigned higher priority than storage and preservation of
history.
5.4.2 Security Perspective
In a fog computing infrastructure, end-to-end security must cover

everything between the cloud and the things on the edge of the network. In
the architecture, security starts with the individual fog node hardware. If the
node is not designed with the appropriate security to ensure that it is a
trusted element, it isn’t possible to build a trustworthy end-to-end fog
computing infrastructure. Once trusted fog nodes have been deployed, a
secure fog network can be layered on top of the node infrastructure,
providing the basis for secure node-to-node, node-to-thing, and
node-to-cloud communication.
5.4.2.1 Trusted and Trustworthiness
Trustworthy fog systems depend on using trusted elements that are

responsible for maintaining the security policies specified for a given device.
If one or more trusted components (hardware, firmware, or software) in a
node are compromised, then the node—and, by extension, the system—is no
longer trustworthy. The RA also determines trustworthy attributes including
behavior by inspection of historical behavior at various levels of the
hierarchy to determine if the system and its components are acting in a
trustworthy manner.
Policies specify who or what can access which resource under what
circumstances. Security mechanisms then implement the policy. Some
policies may be embedded in the node’s hardware and software. Other
policies may be pushed from the fog management system to the node; they
can be added, changed, or deleted by an authorized management or
orchestrator administrator. Each layer in a fog deployment requires security
as represented in the figure below.
47
Figure 15 OpenFog Security Layers
5.4.2.2 Threat Models and Threat-Based Design
Fog deployments require security protection mechanisms implemented in a

particular design dependent on the threat model and the value of the asset
being protected for that fog node. In the architecture, we assume that
attackers are actively looking to compromise the assets, looking for the most
vulnerable entry point. The objective is to provide sufficient security for the
threat model and upgrade the security, as needed, over time. Error! R
eference source not found. lists some (non-exhaustive) examples of
threats and attacks towards fog nodes. When designing for threats in a fog
environment you need to understand the various views of each and the
overall deployment model that is being addressed. In many fog
deployments, you cannot assume physical possession is out of scope and
that further adds requirements onto fog platforms.
Different use case models, even within a single use case vertical, may
require different types and levels of security. Many different types of assets
need to be protected against different levels of threats specific to their
intended use and location. Assets may include:
 Information Technology infrastructure

 Critical infrastructure
 Intellectual property
 Financial data
 Service availability
 Productivity
 Sensitive information
 Personal information
 Reputation
48
Threat: An action (attack) that can damage an asset or cause a security
breach.
Threat model: Specifies the types of threats that the system defends
against, as well as threats that are not considered. A threat model should
clearly specify what assumptions are being made about the system, its
users, and potential attackers. A threat model need not describe the details
of the attacks that it protects against. It should specify whether attacks on
the operational system in the field are the only ones considered, or considers
attacks during the development by an insider. Insider attacks are typically
much harder to protect against, because the designer can build in a back
door that can be exploited later.
Threat Confidentiality Integrity Authentication Availability Privacy

Categories Violation Violation Violation Violation Violation
Intents Leaking Modifying Masquerading Rendering Leaking

information data/code one entity as resources sensitive
through without another entity unreachable information
Attack
overt/covert proper /unavailable of an entity
Venues channels authorization (incl. identity)
Insider Data Identity/Password Equipment Data/Identity

Data Leaks
Attacks Alteration / Key Leaks Sabotage Leaks
Hardware Radio Hardware
Hardware Trojans, Hardware Jamming, Trojans,
Hardware Trojans
Attacks Side Channel Trojans Bandwidth Side Channel
Attacks Exhaustion Attacks
Malware,
DoS/DDoS,
Software Social
Malware Malware Malware Resource
Attacks Network
Depletion
Analyses
Network Message / Spoofing, DoS/DDoS,
Traffic Pattern
Based Eavesdropping Transaction Man-in-Middle Subnet
Analyses
Attacks Replay Attacks Flooding
Figure 16 Example Threats and Attacks
5.4.2.3 Confidentiality, Integrity, and Availability
There are three cornerstones of the fog security perspective:
Confidentiality: The prevention of the disclosure of secret or sensitive

information to unauthorized entities.
49
Integrity: The prevention of unauthorized modification of protected data or
code without detection.
Availability: The ability of a system to continue to provide service to

authorized entities at the agreed upon service level as needed. Availability,
in the security sense, has to consider external attacks, such as Denial of
Service, rather than just hardware and software failures and faults.
5.4.2.4 Access Control
Restricting access to resources (objects) to only those allowed to access that

information is key to building a secure system. Access Control encompasses
Authentication, Authorization, and Accounting (AAA).
 Authentication answers the question “Who are you?” Authentication

is used between humans and machines and between machines and
machines.
 Authorization answers the question “What are you allowed to do?”
 Accounting refers to the record keeping and tracking mechanisms
implemented in the system. This includes tracking and logging access
to system resources.
 Physical access security to insure only authorized people are allowed
to touch fog hardware is another aspect of this security mechanism.
5.4.2.5 Privacy
Privacy is the right to decide how one’s information is used. (Confidentiality

is the obligation to protect secret or sensitive information.) Privacy is a
property of data. Fog systems must allow users to specify the privacy
attributes of the data that they own on the system. In a multi-tenant
system, this may involve specifying both privacy and sharing rights among
tenants. If fog systems capture data for analysis at the edge, the privacy of
that data must also be accounted for in the deployment.
5.4.2.6 Identity and Identity Protection
Public-key ciphers can be used for establishing a longer-term cyber identity,

e.g., for authentication. In public-key cryptography, keys come in matched
pairs (public key and private key) for each user, entity, computer, or
subject. The private key must be accessible only to the subject and
represents the subject’s digital identity in cyberspace.
Hashes can be used to verify the integrity of code modules by taking the
hash of the good known code module and using that to identify the module
50
(like a unique global name). The same code module infected with malware
would have a different identity or hash value. Two identical code modules or
data, but with different filenames, would have the same hash value, which
means the same identity.
The private key of someone’s key pair is like their digital identity. For
example, an operation executed with Alice’s private key can authenticate the
person as Alice. Private keys must be kept confidential in order to protect
someone’s digital identity.
5.4.3 Manageability Perspective

Many fog computing deployments involve machine vision, and associated
human-like functions. As such, they have the ability to see, respond,
remember, move, and make autonomous decisions in order to participate in
other fog services. This range of actions requires a higher level of
manageability than a traditional static model. In addition, fog nodes may be
deployed in a wide range of locations: remote, fixed and non-fixed, and
environmentally harsh conditions.
Fog computing is driving changes to manageability service compared with in
traditional IT and OT management systems.
5.4.3.1 Manageability Interfaces

Manageability interfaces deployed for fog computing should support In Band
(IB) or Out of Band (OOB) management interfaces or both. There are pros
and cons for using IB or OOB manageability, but the best choice is generally
dependent on the given deployment scenario. However, both may be used
as we move to autonomous levels of manageability.
 IB manageability: This refers to the manageability interfaces that
are visible to the software and firmware running on a given system. IB
manageability interfaces may communicate with a system service
processor (SSP) or a baseboard management controller (BMC), if they
are present on a given hardware node. However, this is not required.
In some scenarios, IB manageability may be run on a separate OS
thread or periodic service. Many systems use “heartbeats” to manage
the health of a given system. If the IB management thread does not
send the heartbeat, a higher-level management entity may restart or
alert service systems to address the issue.
 OOB manageability: This refers to manageability interfaces to a
manageability subsystem that is not running on the host operating
system. These are generally discrete manageability systems that can
survive and manage systems in all power states. Specifically, when a
fog platform is powered off, and software is not executing on the host
51
platform, an OOB manageability interface can still be used
communicate with the platform and perform things like inventory
control, system health, power it one, etc. Examples include BMCs as
defined by the IPMI specification. OOB management has potential
security advantages, especially for business-critical IoT applications.
5.4.3.2 Management Lifecycle

Even the smallest fog node has a management lifecycle. The figure below
shows the main components of a management lifecycle from a
manageability perspective.
In all systems, there are one or more management agents. These may be
implemented as discrete systems or software services. The purpose of the
management agent is to ensure that each element of a fog node successfully
goes through the management lifecycle. Automation is important during all
phases of the lifecycle, because human intervention is impractical for large
fog networks.
Figure 17 Management Lifecycle
Commission: This is the earliest phase of a fog platforms lifecycle. When a

managed entity is commissioned, certain actions are required prior to
provisioning. These include identification, certificates, calibration of time,
etc. In addition, at this state the managed entity must:
 Include security that can be attested to and trusted in future phases of
the lifecycle.
 Include RAS (reliability, availability, and serviceability).
 Be agile in data collection and monitoring.
 Open in order to allow control and provide visibility into its resources.
Provision: When a managed entity begins its early life in a fog node it must
be enrolled. This includes discovery, identifications, advertisement of
features and capabilities, trust, and deployment of features. The managed
52
entity must also be scalable upon provisioning. It must have the ability to
support a multitude of hierarchies.
Operate: When a fog node is in normal operation. Manageability
requirements cover all aspects of reliability, availability, and serviceability.
Recovery: When a fog node is operating out of expected norms, it must be
autonomous in its ability to recover. It should attempt to self-heal and
perform recovery operations. Other fog nodes may also assist with the
recovery action, which is why the architecture defines both OOB and IB
manageability interfaces.
De-Commission: Since many aspects of fog nodes may have Personally
Identifiable Information (PII), the architecture specifies an ability for
cleansing all aspects of hardware. This includes the ability to decommission
fog node instances and re-use them for another deployment. It includes
ways to securely wipe out all of the Non-Volatile (NV) storage so that future
applications may not access the previous tenant’s data.
5.4.3.3 Management Layer

As shown in management lifecycle, the fog management layer has many
responsibilities, including automated discovery, registration, and
provisioning of endpoint devices. Discovery services provide an efficient
method for finding, identifying, onboarding, and managing components in
the fog infrastructure. Both IB and OOB discovery methods are used. IB
services are usually discovered using operating system or software agents.
OOB discovery is usually performed through wireless, SMBus, or I2C
interfaces which are easier to maintain during low power system states. The
purpose of discovery is to gain a full understanding of the endpoint device’s
resources, establish a health baseline, and ensure correct operational state
until the element is decommissioned.
53
Figure 18 Management Layer
The most commonly used manageability aspects of a fog node are system
software and firmware update and remote alerts of abnormal system
operation. As fog-based systems often operate in harsh or remote
environmental conditions, it is usually a requirement to provide “over the
air” (OTA) firmware and software updates. The manageability layer is
responsible for these updates.
5.4.4 Data, Analytics, and Control

The traditional way of delivering analytics is no longer efficient or, in some
cases, even possible using traditional sensor to cloud models. This is due to
the large volume of data that must be captured, stored, and transported to
the data center or cloud for additional processing and analysis by large-scale
business applications. As we look deeper into business and technical
processes, more granular data elements will be needed to create actionable
business knowledge from information. This data journey is an evolving vision
for most companies and institutions. Some are interested in taking the full
journey. Some just want to understand the current state of their operations
and want Descriptive Analytics (the analysis of what’s happening or what
happened). Others are interested in Diagnostic Analytics, which entails root-
cause analysis. Predictive Analytics uses all the knowledge from the previous
54
analysis and combines it with other knowledge about processes and tools to
understand what will happen. Eventually companies may be interested in
Prescriptive Analytics, which enables processes to optimize themselves.
The more that companies want to understand about their operations, the
more data, compute, and data resources they will need. As we have shown
multiple times, with fog computing, we have the tools to capture, store,
analyze, and transport only relevant data. This is done by having the
intelligence of upper layer data-center or cloud applications embedded as
close as possible to the data-source. This means that with integration
between the data-source and the Business Intelligence analytics
applications, the “network” or the “edge” will capture the data from the
source, process it for local purpose-specific analytics, deliver an action back
to the process while at the same time sending same or other datasets to the
Data Center or Cloud for further “business or operations-specific” processing.
The hierarchical nature of fog helps this, allowing different components of
the analytics algorithms to operate at different fog layers. We see this as
cross cutting concern because it has to happen at the right layer as dictated
by the scenario being addressed.
Figure 19 Business Intelligence
The figure above shows the integration of and open data exchange across all
elements of the business process. This integration and exchange is
necessary for the success and accuracy of business intelligence analysis. The
OpenFog Consortium is working on developing (and evolving) various
solutions around security and identity to facilitate this type of
communication within an enterprise and between the enterprise and its
partners and suppliers. Business intelligence will depend on well-defined
flows, secure boundaries, facilitating data capture and exchange among the
various data processing elements, and the data science capable of making
sense of it all.
55
5.4.5 IT Business and Cross-fog Applications
Fog applications and services should have the flexibility to span and
interoperate with various levels in a fog hierarchy. This is a foundational
aspect of fog computing that enables a multi-vendor ecosystem. In addition,
the data that is collected or generated by one fog node should be shareable
with other nodes in the hierarchy. The figure below shows a cross-fog
application spanning east to west nodes (it would also be able to span north
to south).
Figure 20 Cross Fog Application
Cross-fog applications require that we have an understanding and adoption

of smart objects and associated data models. These are critical to those
applications for interoperability and additional value creation.
Node View
As previously described, the OpenFog RA description is a composite of
multiple stakeholder views. The node view is the lowest level view we
currently utilize in the architectural description. The stakeholders involved in
formulating the viewpoint (and subsequently this view) are the system on a
56
chip designers, silicon manufacturers, firmware architects, and system
architects.
Figure 21 Node View
Before bringing a node into a fog computing network, the following aspects
should be addressed:
Node Security: As described previously, node security is essential to the
overall security of the system. This includes protection for interfaces,
compute etc. In many cases a node will act as a gateway for legacy sensors
and actuators to higher-level fog functions and therefore can act as a
security gateway. It is important to note that Node security is shown as both
a vertical perspective as well as a horizontal requirement for this view. This
is an important concept as security must be considered at all levels from
silicon to software.
Node management: A node should support the management interfaces,
provided by the node being managed. Management interfaces enable higher-
level system management agents to see and control the lowest level node
silicon. The same management protocol can be used across many different
physical interfaces.
Network: Every fog node must be able to communicate through the
network. Since many fog applications are time sensitive and time aware,
some fog computing networks may need to support Time Sensitive
Networking (TSN).
Accelerators: Many fog applications utilize accelerators to satisfy both
latency and power constraints as it relates to a given scenario.
Compute: A node should have general purpose compute capabilities. It is
also important that standard software (e.g., Commercial off the Shelf or
open source) be able to run on this node. This enables a higher level of
interoperability between fog nodes.
Storage: An autonomous node must be able to learn. Before any learning is
possible, it must have the ability to store data. Storage devices attached or
embedded to this node need to meet the required performance, reliability,
and data integrity requirements of the system and scenario. In addition,
57
storage devices should also provide information and early warnings about
the health of the media, support self-healing properties, and support ID-
based performance allocation. Some kind of local, standalone storage will be
required for local context data, logging, code images, and to service
applications that run on the node. There will often be more than one kind of
storage required – e.g., local hard disk, SSD, and secure storage for keys
and other secret material.
Sensors, Actuators, and Control: These hardware or software-based

devices are considered the lowest level elements in IoT. There could be
several hundred or more of these associated with a single fog node. Some of
these are dumb devices without any significant processing capability, while
others may have some basic fog functions. These elements generally have
some amount of connectivity, and include wired or wireless protocols, such
as I2C, GPIO, SPI, BTLE, ZigBee, USB, and Ethernet, etc.
Protocol Abstraction Layer: Many of the sensors and actuators on the
market today are not capable of interfacing directly with a fog node. The
protocol abstraction layer makes it logically possible to bring these elements
under the supervision of a fog node so that their data can be utilized for
analytics and higher level system and software functions.
Abstraction is also key to multi-vendor interoperability for both IoT things

and fog nodes. Well-known inter-element interfaces supply a layer of
abstraction. They enable vendors to share meta-data of fog architectural
elements they support, which in turn fosters multi-vendor data
interoperability and service composability. When meta-data is exposed, it
also may be used for cross-layer optimizations, for example to optimally
route data between fog nodes using information-centric networks (ICN) or to
create dynamic fog topologies as with software-defined networks (SDN).
Future versions of the OpenFog RA will describe “Minimum Viable

Interfaces”, which will include many more details about protocols and
abstraction layers. Following subsections contain more details of the aspects
mentioned above, except those of the sensors, actuators, controls and their
internet protocol abstractions. Please refer to the security appendix for
further discussions on IoT device connectivity and security issues.
5.5.1 Network
Fog nodes generally are of most value in scenarios where data needs to be
collected at the edge and where the data from thousands or even millions of
devices is analyzed and acted upon in micro and milliseconds. In these
scenarios, various networks facilitate communication within the fog nodes to
58
sensors and up to higher levels in the hierarchy up to and including the
cloud.
The network should provide the scalability, availability, and flexibility

required by the communication pattern or process. The network should also
provide whatever QoS is required to prioritize critical or latency-sensitive
data and even guarantee delivery. QoS should be addressed at the lowest
level, hence Node view, to provide guarantees of QoS. This enables higher
level views of system providers (System view) and applications (Software
view) to build upon a stable foundation.
In the following sections, we will explore various network elements from the
point-of-view of a fog node’s connectivity and communications requirements.
Note: Depending upon the deployment scenario, fog nodes will most likely
exist within a network element, such as an access point, gateway, or router.
In the architecture, we assume that the network requirements will be the
same regardless of the placement of the fog node. This clearly will change
based upon deployment and we will refine this through Testbed and other
open deployments.
5.5.1.1 Wired Connectivity
The network connectivity model for a fog node will depend on the node’s
purpose and location. For example:
 A fog node in a factory used to gather and analyze manufacturing

process data will most likely be connected to upper and lower layers
using a wired network.
 A fog node used for gathering and analyzing personnel locations will
most likely be connected to the sensor using a wireless network.
 Internal connections within fog nodes will most likely be connected
using RDMA and other low latency interconnect technologies.
There are many standards, types, and interfaces for physical connectivity
that can be utilized to connect a fog node. Physical connectivity is typically
one or more Ethernet links supporting speeds ranging from 10 Mbps to 100
Gbps, supported on copper or fiber links to serve various reach
requirements. We typically see copper connectivity used for links up to 100
meters in length and supporting speeds ranging from 10 Mbps up to 1 Gbps.
For connections requiring higher speeds and longer reach, optical fiber
cables may be used. Optical fiber support different wavelengths and
transmission modes to support the required distances and capacities. For
59
example, a single mode optical fiber cable (utilizing a single ray of light)
supports longer distances than a multimode fiber (utilizing multiple rays and
multiple wavelengths).
For connecting a fog node to IoT devices or sensors, there are a variety of
standards and interfaces that are vertical or solution dependent (non-
Ethernet protocols). For example, in an industrial environment, the fog node
may be required to support a Controller Area Network (CAN) bus or other
fieldbus standards for communicating with lower layer applications and
processes.
For industrial automation uses, guaranteed data delivery is critical. This type
of networking (usually using Ethernet) is called Time Sensitive Networking
(TSN) also known as Deterministic Ethernet. TSN uses standards-based time
synchronization technology (e.g., IEEE 1588) and bandwidth reservation
(class-based QoS) to prioritize control traffic in a standard Ethernet
environment.
If a fog node is required to interface with devices in industrial automation,

automotive, or robotic environments over Ethernet, support for TSN may be
needed.
5.5.1.2 Wireless Connectivity
Wireless connectivity is an essential part of the Internet of Things in

particular, and the digital transformation in general. Wireless connectivity
provides flexibility and enhances efficiency and productivity. Wireless
interfaces come in a variety of protocols, standards, and mechanisms. The
quality of connectivity is dependent on many conditions including but not
limited to the level of flexibility, mobility, reach, availability, power
constraints, and environmental or geographical conditions. For the various
IoT applications envisioned for fog computing, wireless connectivity seems to
be especially beneficial for southbound communication (sensor-to-fog node),
but will also be used for fog node-to-fog node and fog-to-cloud
interconnects.
Wireless support at the fog node will be dependent on a variety of

parameters, including:
 Function and/or position in the hierarchy.

 Mobility is an essential property to support in-vehicle fog nodes and
many classes of geographically-distributed IoT endpoints. Wireless
interfaces are the only practical way to reach them.
 Coverage required to satisfy the deployment requirements.
60
 Data volume (throughput) and velocity (data transfer rates).
 The form factor required to support various types of antennas,
modules, or transceivers.
 If the fog node is expected to receive, process, and relay information
upstream to other layers in the hierarchy at a high rate, then the
energy source (including efficiency, delivery, and dissipation
considerations) becomes a significant design factor. Similarly, if
transmission and processing rates are lower, then battery, harvested
energy, and rechargeable sources may be used to fulfill design
parameters.
 Wireless support is highly dependent on the environment, i.e.,
interference. For example, deploying wireless technologies in noisy
environments or around highly reflective metallic surfaces may affect
the performance required of the fog node. This is also true for
maritime and other areas where certification of physical infrastructure
requires that structural integrity of a fog node is not compromised by
antenna attachments/cable attachments. In these scenarios, secure
and reliable wireless communication is a requirement.
 Using a licensed spectrum normally requires a fee for accessing the
frequency ranges, while, in most cases, using an unlicensed spectrum
may be low or no cost.
Wireless connectivity can be grouped into three major areas: Wireless WAN
(WWAN), Wireless LAN (WLAN), and Wireless Personal Area Networks
(WPAN).
Note: It is not uncommon to refer to some of the types of communication

technologies as Wireless Metropolitan Area Networks (WMAN). For the
purpose of this discussion, and due to the interchangeable nature of WWAN
and WMAN use-cases, we use WWAN as a superset of both. Wireless WAN
(WWAN): WWAN technologies are used when large geographic area
coverage is required. A variety of protocols and standards exist and we list
the following standards a fog node may be required to support to
communicate with the other devices or nodes within the same network:
 Cellular technologies, including 3G, 4G LTE, and 5G, feature high data
transfer rates (>1Gbps). They also cost more (as a licensed spectrum)
and are less power efficient. The majority of cellular communication is
standardized by the Third Generation Partnership Project (3GPP).
o Note: 5G promises higher speeds, higher capacity, and much
lower latency than the current cellular systems. With 5G it may
be possible to deliver double-digit Gbps speeds (10+ Gbps). 5G
promises to facilitate higher adoption of IoT solutions and
61
devices. Fog nodes aggregating data from cars, mobile devices,
and sensors will most likely need to support 5G.
 Fog nodes may be required to support cellular technologies for
southbound or northbound traffic or both. Fog nodes may use cellular
technology to back-haul sensor traffic. Most types of mobile fog nodes
will use cellular northbound interfaces. Many software defined radio
technologies that physically connect into the fog node address some of
these scenario requirements.
 Narrow Band IoT (NB-IoT) is a 3GPP standard that addresses a variety
of IoT applications and requirements and delivers on the promise of
long-range and lower power requirements. NB-IoT is not widely
available yet.
 Low-Power Wide Area Networks (LPWAN) have low data transfer rates,
higher power efficiency, and are low cost. Proprietary LPWAN
implementations are being tested by various organizations, such as
the LoRa Alliance and Sigfox. LPWANs are currently being investigated
for agriculture applications because of their ability to cover large areas
of farming and rural lands.
Wireless LAN (WLAN): WLANs utilize a variety of topologies and protocols,

but WLAN has become synonymous with WiFi. WLANs are a good
communication choice for smaller geographical areas, often within a building
or campus. Depending on the number of access points and the density
requirements, WLANs may also be used in stadiums, manufacturing plants,
and oil and gas refineries and fields. The following are some examples of
WLANs that may be supported by the fog node:
 WiFi (WLAN) is defined by a group of IEEE 802.11 standards. These

standards address multiple requirements and challenges for the
environment in which they are deployed. They support data transfer
rates from few Mbps up to multiple Gbps. The most common standards
are IEEE 802.11a, b, g, n, and ac. IEEE802.11ac is the latest in a
series of evolutions of the IEEE 802.11 standards; it supports higher
densities and transfer rates.
 The IEEE 802.11 workgroups are currently working on new solutions to
address the need for higher capacity, density, and speed, especially for
IoT use cases. For example, the IEEE 802.11ax is expected to bring
higher speeds and capacities beyond the capabilities of 802.11ac. The
802.11ah is tailored for IoT use cases requiring low power
consumption and longer ranges. IEEE 802.11p will define standards for
Vehicle-to-Vehicle as well as Vehicle-to-Roadside infrastructure
communication.
 Future developments in free space optical communications such as Li-
Fi hold promise as options for fog wireless networking.
62
Wireless Personal Area Networks (WPAN): WPAN is characterized by a
short communication range, low power consumption, and low cost. WPANs
may be used with wearable devices and home management systems. WPAN
includes the following technologies:
 Bluetooth: Characterized by short-range communication. Specifications

and standards managed by Bluetooth Special Interest Group (SIG).
 Infrared (IR): Characterized by line of sight wireless communication
over IR light waves. Specification provided by IrDA (Infrared Data
Association).
 ZigBee: Characterized by low power consumption, short range (up to
100 meters, given the right environmental conditions), and low data
transfer rates.
 Z-Wave: Characterized by RF signaling and control mostly used in
home automation.
 IEEE 802.15.4 (Low Rate WPAN) also applies to WLAN use cases. It is
also the standard that defines Layers 1&2 of the OSI model.
Near Field Communication (NFC): NFC is a technology that may be used

when fog nodes support devices that need to communicate in very close
proximity. NFC technologies have been used in logistics and supply chain
solutions for quite some time; now they’re being used in vertical markets,
including frictionless solutions for retail, agriculture, and healthcare.
Solutions like Passive RFID, which also utilizes NFC, are used for asset
tracking and physical access.
Wireless connectivity to the fog node allows the ability of various sensors
and data to flow into the node where it will be processed. As node
capabilities grow, it enables higher level secure communication functions to
be utilized in the architecture.
5.5.1.3 Network Management
As the number of sensors and data sources increase, the need to manage all
assets, nodes, and resources increases in importance. The ability of fog
nodes to be supported by out-of-band (OOB) network management will help
manage resources, security, health of the fog node, and the ability to adapt
to changing conditions in the environment. The protocols and mechanisms
available to manage sensors, nodes, and network devices vary based on the
communication protocol used and the availability of connectivity options and
CPU/memory resources. In some cases these are separate managed
networks; in other cases the management communication is sent on the
host network. There are multiple ways information is provided to the node,
63
and to ensure continued secure, reliable, and safe operation, it is important
that these things and associated networks are capable of being managed
with both sophistication and simplicity.
5.5.1.4 Network Based Security Threats and Mitigation
The fog node needs to be protected from various network-based security

threats, which may include:
 Denial of Service attacks

 Intrusion
 DNS spoofing
 ARP spoofing or poisoning
 Buffer overflows
Fog nodes may not always be capable of protecting against these types of
attacks. They will most likely depend on the network or adjacent devices to
help protect them. The following are some of the common examples for
network devices that help protect the fog node:
 Firewalls for blocking unauthorized access.

 Intrusion Prevention Systems (IPS).
 Secure Remote Access using Virtual Private Networks.
 Behavior-based anomaly detection appliances or software.
Several instances of massive DoS attacks that used network attached IoT
devices could have been detected much faster and potentially mitigated by
utilizing fog driven network security.
Please refer to the Section 10.1.3.1 for detailed discussions on the network
and data security aspects of OpenFog architecture.
5.5.1.5 Network Design Considerations for Fog Nodes
Whether a fog node is being installed into an existing brownfield or new

greenfield networked environment, the following design considerations
should be kept in mind:
 Physical or virtual implementation of fog node capabilities depends on

capacity and policy needs. For example, the ownership of the physical
infrastructure could be the factor that determines access to the virtual
elements supported within the infrastructure.
 Fog node-to-node communication considerations:
o Direct or indirect
64
o Distance between any two nodes, which influences energy use,
bandwidth, cable complexity, and cost
o Requirement for state preservation between two nodes (in a
backup or high availability scenario)
o Type of communication interface and protocol
 Capacity planning:
o Start with the end state in mind when architecting for the
scenario
o Understand the impact of new traffic patterns on the fog node
and the network as a whole
 Streaming data readiness:
o Volume of data expected
 The frequency of maintenance and upgrades
 Convergence with IT: Choosing Ethernet and IP for node-to-things
communication may allow for easier integration with the IT
environment and contribute to the efficiency of the system.
5.5.2 Accelerators
In addition to traditional CPUs, some fog nodes, especially those engaged in
enhanced analytics, require CPU throughput in excess of what can be
economically (power or processing efficient) provided by standard current
server and enterprise CPU chips. In these cases, accelerator modules will be
configured next to the processor modules (or tightly integrated) to provide
supplementary computational throughput. Here are some examples:
 Graphics Processing Unit (GPUs) often contain thousands of simple
cores. For applications that can efficiently exploit their massive
parallelism, they can be an order of magnitude faster and provide
significant power and space savings. Multiple GPUs can be equipped on
each standard CPU. However, to achieve this capability power delivery
and physical and electrical connection to the node would need to be
increased which can increase the overall node power consumption.
 Field Programmable Gate Arrays (FPGAs) are large collections of gate-
level programmable hardware resources. They can be configured with
custom logic designs to solve very specific problems very efficiently.
However, depending upon deployment the additional power reduction
compared with other accelerators may require additional lower level
knowledge (e.g. VHDL). In many cases FPGA provide better power
efficiency compared with discrete GPUs.
 DSPs are specialized processors optimized to operate on signals. Some
DSPs are general purpose, while others are optimized for special
functions, like video compression and manipulation.
65
When choosing an accelerator for a given node, you must balance the
following:
 Interfaces to the node (electrical throughput and power delivery).
 General applicability to multiple scenarios. This means the dynamic
ability to change (programmability pillar) the accelerator to address a
new use case.
 Dynamically changing requirements.
 Environmental constraints.
 Higher level software programming interfaces and API support,
especially the orchestration needed to discover an accelerator’s
existence and capabilities
The programmability pillar implies the node’s ability to be flexible in how it is
defined to address a given problem. This late binding is especially important
in many of the areas of interest to the OpenFog consortium.
5.5.3 Compute
As more and more data is processed at the edge, this will increase the
computational requirements at the true edge of the network. Having general
purpose computation at the edge will continue to be important. Additionally,
larger amounts of system DRAM paired with this compute will also grow in
importance. To ensure higher reliability and accuracy of computation we will
start to see ECC memory start to become more prominent in the compute.
Another key importance of fog computing is that deployment lifecycles may
be longer so it is important that long life and compute performance is
calculated to be sufficient throughout a deployments lifecycle. Compute is
likely to be implemented as one or more multi-core CPU. While other
architectures exist, it is important to balance the programmability aspects
when considering different architectures.
When considering a compute element, it is important to understand the

environmental conditions it must correctly operate under. In many fog
deployments, the compute must continue to operate well beyond 70 degrees
C. In fact, the harshest environments of up to 100 degrees C is not
uncommon in many parts of the world.
The compute function embodies many requirements for fog nodes. Following
are some examples of computational requirements:
 Some multi-tenant installations will dedicate whole cores to each of

their most critical applications. This capability may be required to
ensure QoS.
66
 Memory management units may be required in fog to manage large
virtual memory space, to isolate platform from application space, and
to isolate applications from each other in multi-tenant environments.
 High performance I/O subsystems are required in fog in order to
connect each CPU with its associated accelerators, storage, and
network peripherals.
 In some fog node designs, the hardware root of trust is located within
the CPU complex itself and verification of code only happens after the
CPU verifies the signature.
5.5.4 Storage
Many types of storage will be required in fog nodes. As fog computing

continues to emerge we will see storage tiers typically only seen in
datacenters emerge on nodes as they collect and process this data across
the hierarchy. This includes:
 RAM Arrays: As data is created from sensors, the node will need to
operate on that data in close to real time operation. RAM arrays satisfy
this requirement versus additional latency when accessing non-volatile
storage. Many fog nodes will also have on-package memory to satisfy
the latency aspects required for certain scenarios.
 Solid State Drives: Flash-based storage may be used for the majority
of fog applications because of its reliability, IOPs, low power
requirements, and environmental robustness. These include PCIe and
SATA attached SSDs. Additionally, new classes of solid state media are
beginning to emerge with new programming models. These include
3DXpoint and NVDIMM-P.
 Fixed Spinning Disks: For large, cost sensitive storage applications,
fog nodes may contain rotating disks, sometimes arranged as
Redundant Array of Inexpensive Disks (RAID) arrays.
The actual storage medium chosen depends upon the use case; within a
given fog node, there will typically be a hierarchy of storage options.
Ultimately, storage devices need to meet the cost, performance (IOPS,
bandwidth, and latency), reliability, and data integrity requirements of the
system. In addition, storage devices in fog computing need to support the
OpenFog pillars, particularly the security and RAS pillars. However, the
biggest advancements will come as flash based storage technologies
continue the march to cost/byte and access latency times of DRAM.
Storage devices should provide support for encryption and key management
and authentication by supporting standards such as AES-256 and TCG Opal
67
etc. The storage device should also provide real-time information and early
warnings about the health of the media and support self-healing properties.
Finally, in virtualized fog computing environments the storage device should
support ID-based performance allocation by providing adjustable storage
resources (IOPS or bandwidth) to specific applications or virtual machines.
Supporting data encryption at rest is also important in most fog deployments
as they will be deployed in areas where physical protection mechanism seen
in the data center are no longer true.
5.5.5 OpenFog Node Management
OpenFog node management refers to manageability systems that are not

running on the host operating system. These are generally discrete
manageability systems that can survive and manage fog nodes in all power
states. They are also sometimes called Hardware Platform Management
devices (HPM).
Most fog nodes will include a HPM that is responsible for controlling and
monitoring the other components inside the node (e.g. storage, accelerators,
et al). The HPM is typically a small auxiliary processor on the main CPU or
motherboard. It has various sensors and monitoring points to track variables
like temperature, voltage, current, and various errors. These readings may
periodically be reported to external RAS system. If serious errors are
detected, alarm notifications can be escalated by the HPM subsystem.
The HPM system is also responsible for controlling the internal configuration
of fog nodes. It may set communication parameters like IP addresses and
line speeds. It can configure new hardware modules. If modules fail, it can
isolate them and attempt to recover their functions. The HPM subsystem also
cooperates with a trusted component in the chain of of trust to securely
download software updates for the entire node. Sensors associated with the
physical operation of fog node hardware, including ambient temperature,
airflow, fan speed (if used), supply voltage, supply current, moisture,
cabinet door tamper, etc., also send data through the HPM subsystem.
In many deployments because the HPM can operate out of band with the
main computational elements, it too will have its own TPM and go through a
secure boot process. The HPM just like other entities should support a HW
root of trust.
68
5.5.6 OpenFog Node Security
As described in the Security Perspective, it is important to perform a security

analysis and threat assessment of the fog implementation in order to
properly identify the needs the fog node. Once this task is accomplished, you
should have the information needed to determine the appropriate physical
security measures, the optimal method for establishing and maintaining
trust, and what type of policies to put in place in order for the fog node to
securely manage and respond to its environment.
5.5.6.1 Physical Security and Anti-tamper Mechanisms
The level of physical security supported by a fog node should be aligned with
the security policy for that device and threat level. This will depend on how
difficult it is to access system components and what the consequences are if
the system is breached. The location of the fog node and the degree of
physical access available in that location will play a role in the evaluation.
Fog nodes located in open public spaces such as shopping malls, street
corners, utility poles, and even in a personal vehicle will provide greater
opportunity for a physical attack. Note: There may be industry specific
standards and requirements for providing physical security for these devices.
These are not addressed here.
The goal of anti-tamper mechanisms is to prevent any attempt by an

attacker to perform an unauthorized physical or electronic attack against the
device. Anti-tamper mechanisms can be divided into four groups:
 Resistance
 Evidence
 Detection
 Response
It is important that legitimate maintenance measures should not damage the

node due to anti-tamper mechanisms. To help prevent this, the node might
have a special maintenance mode that can be configured by an authorized
entity so that the tamper response is disabled while maintenance is in
progress and then re-enabled.
5.5.6.2 Tamper Resistance
Tamper resistance uses specialized physical construction materials to make

tampering with a fog node difficult. This can include such features as
hardened steel enclosures, locks, encapsulation, or security screws.
Implementing tight airflow channels (that is, tightly packing the components
69
and circuit boards within the enclosure) will increase the difficulty of using
fiber optics to probe inside the node without opening the enclosure. When
we look at the node view, this includes all interfaces enabled by the SoC
manufacturer. Many manufacturers enable special modes of operation called
manufacturing or test modes. These modes of operation must be protected
from tampering from physical attack after the node is deployed.
5.5.6.3 Tamper Evidence
The goal of tamper evidence is to ensure that, when tampering occurs,

visible evidence is left behind. Tamper evidence mechanisms are a major
deterrent for minimal risk takers (e.g., non-determined attackers). Many
kinds of tamper evidence materials and devices are available, such as
special seals and tapes that make it obvious when there has been physical
tampering. In the previous example, tampering could notify the HPM to
ensure higher level management entities can determine tampering without
physically being present.
5.5.6.4 Tamper Detection
Tamper detection means that the system is made aware of unwanted

physical access. The mechanisms used to detect the intrusion typically fall
into one of three groups:
 Switches, such as micro switches, magnetic switches, mercury

switches, and pressure contacts, detect the opening of a device, the
breach of a physical security boundary, or the movement of a
particular component.
 Sensors, such as temperature and radiation sensors, detect
environmental changes. Voltage and power sensors may detect glitch
attacks. Ion beams may be used for advanced attacks to focus on
specific electrical gates within an integrated circuit.
 Circuitry, such as flexible circuitry, nichrome wire, and fiber optics
wrapped around critical circuitry or specific components on the board,
is used to detect a puncture, break, or attempted modification of the
wrapper. For example, if the resistance of the nichrome wire changes
or the light power traveling through the optical cable decreases, the
system assumes there has been physical tampering.
 Mesh enclosures, such as Gore’s Tamper Responsive Surface
Enclosure, are designed to protect the physical security boundary of a
fog node and combine a number of tamper evidence and detection
features.
70
All of these tamper detection mechanisms typically provide a hardware
security violation signal to a security monitor when they are tripped.
5.5.6.5 Tamper Response
Tamper response mechanisms are the countermeasures taken when

tampering is detected. The response to such an event should be
configurable.
 Soft Fail: Sensitive data is cleared and a second interrupt signal is

sent to the security monitor to confirm this has been done so that it
can restart the processor and continue execution.
 Hard Fail: The actions for a Soft Fail are performed, plus the caches
and memory are zeroed and the system is reset. Both lower and
higher consequences may be available. The lowest consequence would
be to do nothing, or the event can be logged for later analysis. An
example of a higher consequence might be “bricking” the device. This
means that after zeroing all sensitive data, caches, and memory, the
node cannot be booted again and must be replaced.
The response to tampering needs to be understood and planned for by

higher levels in the fog architecture. This dependency is often overlooked in
system deployments and therefor an attack surface commonly exploited.
Thus, response of Nodes must be understood by the System, and the
Software running on them.
5.5.6.6 Establishing and Maintaining Trust
5.5.6.6.1 Trusted Computing Base

The Trusted Computing Base (TCB) refers to the platform hardware,
software, and networking components that, if violated, would compromise
the ability of the system to enforce its security policies. The more
components and code that are in a TCB, the harder it is to guarantee that it
is free of bugs and security vulnerabilities. It is desirable to ensure that the
TCB is as small as possible to minimize its attack surface. Sometimes
however, this is not achievable due to the complexity of the system required
to satisfy a particular use case. Creating multiple isolated and protected
regions from the rest of the system is one way of creating smaller TCBs to
reduce the attack surface within a complex system environment.
5.5.6.7 Hardware Root-of-Trust

At the heart of the TCB and the security of the fog node is the root of trust.
There must be no opportunity for malicious actors to highjack the early
initialization or boot processes. Security needs to be anchored in the
71
hardware so that it cannot be circumvented. The Hardware Root of Trust
(HW-RoT) is the key to a fog node’s TCB.
5.5.6.7.1 Secure or Verified Boot (HW-RoT for Verification)
Minimally, fog nodes should support a HW-RoT for verification of the boot
process. Secure or verified boot is an architecture for loading and verifying
signed firmware images, boot loaders, kernels, and modules. It is important
to note that this does not necessarily make use of a TPM even if one is
present.
There are many implementations of verified boot. The process begins
execution with code loaded from an immutable Read Only Memory (ROM);
the compute entity in the fog node may only operate upon cryptographically
signed images. Secure boot implementations may be proprietary in nature.
It is recommended that system architects verify the capabilities and validate
the security strengths of a verified boot implementation. There should be no
mechanism to circumvent the signature method. Additionally, it is important
that the HW-RoT not execute non-verified code. This includes option ROMs
from PCIe devices and other elements that comprise the node view.
5.5.6.7.2 Trusted or Measured Boot (HW-RoT for Measurement)
Trusted booting is different from secure booting because higher level

software can attest (programmatically verify) that firmware running is
secure. One example approach, as described by TCG, defines methods to
perform this function using a TPM. As code executes, it creates a
cryptographic digest of the code modules that are stored in the TPM. The
TPM term used for the storage space for each of these chains is Platform
Configuration Register (PCR). A PCR can be thought of as a single trust chain
used for some specific purpose. This is only one example of many possible
implementations. Other implementations exist while still others may be
described by future innovations. As for verified boot, it is recommended that
system architects verify the capabilities and validate the security strengths
of a measured boot implementation.
5.5.6.7.3 Securing the Boot Process
It is critical that the fog node has a method to securely establish a root of
trust, and that trust is authenticated and extended through the remainder of
the boot process in order to ensure that the node can be trusted.
A fog node must provide a method for ensuring that the firmware and
system software have not been tampered with before components are
executed. There are various methods in which this can be achieved. The
selected and implemented solution to this requirement should be in
72
agreement with your organization’s findings during the security analysis and
threat assessment.
5.5.6.7.4 Identification
A fog node must be able to identify itself to other entities within the
network, and entities that request services from the fog node must be able
to identify themselves to the fog node. The best method for this
identification is to have an immutable identifier with attestation. Attestation
is the ability of a system to provide some non-forgeable evidence to a
remote third-party verifier. One such approach that can enable fog systems
to verify or attest to credentials of a given system while preserving privacy
is the use of a Direct Anonymous Attestation (DAA) implementation.
5.5.6.7.5 Attestation
The security of systems that employ trusted processors depends on

attestation (often referred to as remote attestation or software attestation).
In a fog computing hierarchy, a remote attestation agent can attest to the
authenticity and secure state of a fog system. Depending on the method
used to create the chain of trust (i.e., either measured or verified), and
therefore the trusted environment, the remote agent will attest to different
properties and data. For a system which uses measurement to build the
chain, the remote attestation agent would be able to remotely verify that the
HW-RoT of measurement for a given fog node is correct. In either case, the
objective is to be able to attest to the fact that the firmware and software
running on it are known or trusted. If the firmware running on a given
system fails attestation, it should not be used and remediation should occur.
While there are a number of implementation options available for remote
attestation, the OpenFog Consortium will work with the TCG and other
standards-based approaches for remote attestation across multiple
interfaces.
System Architecture View

The system view of the OpenFog RA is composed of one or more node views
coupled with other components to create a platform. The stakeholders which
typically create these systems to facilitate a fog deployment comprise the
concerns of this viewpoint. Subsequently this view is intended to address the
concerns of the system architects, hardware OEMs, and platform
manufacturers. They also need to understand the Node view such that the
systems they create can be deployed to address a given scenario. The
following diagram shows our visual representation of a system view. We only
show a singular composite image of a node embedded within it, but we also
support the notion of multiple nodes being brought to create a system for
73
deployments that require redundancy or to satisfy other deployment
requirements.
Figure 22 System Architecture View
5.6.1 Hardware Platform Infrastructure
Fog platforms must provide robust mechanical support and protection for
their internal components. This first starts with components selected in the
Node view and extends into the system view. In many deployments, fog
platforms must survive in harsh environmental conditions, as described in
the next section. Some requirements for fog platform enclosures (called the
hardware platform infrastructure in the system architecture) include:
 Compliance with local regulations and standard practices.

 Protection from environmental factors (industrial or commercial
temperature-rated components).
 Resistance to physical attack, vandalism, or theft.
 Acceptable size, power consumption, and weight properties.
 Functional safety requirements to protect people and things from
harm.
 Mechanical support of internal components.
 Management of cooling for internal components.
 Support for node-level modularity and the ability to build and modify
many configurations. This includes the ability to extend the function to
address different deployments.
 Serviceability aspect of a platform.
 Acceptable aesthetics and other factors as fog platforms are deployed
in the world around.
74
5.6.1.1 Environmental Conditions
Many fog platforms will be deployed in harsh environmental conditions. This

means they should have specifications that comply with various international
safety and environmental responsibility standards, such as UL, CSA, ROHS,
and WEEE. Examples of safety and environmental requirements include:
 Temperature ranges required by vertical industries, such as industrial,

automotive, and military. Many systems also cannot operate in
temperatures higher than 60 degrees C. The following temperatures
are generally accepted as norms and represented in Celsius.
o Commercial temperatures: 0 – 70
o Industrial: -40 – 85
o Military: -55 – 125
 Environmental hazards, including humidity, shock, vibration,
contamination, earthquake, and extreme solar load.
 International protection marking (IEC standard 60529) up to the IP 68
level (for example, a waterside node that may be subject to flooding).
5.6.1.2 Thermal
Depending upon deployment, fog platforms for harsh location deployment
may be environmentally sealed. They should not require any fans or other
active elements to maintain safe internal temperatures. They should not
require air filters. However, due to the high-power dissipation and high
packaging density of some higher performance fog nodes, especially those
with large accelerator arrays, active cooling options are allowed. If forced-air
cooling is used, air filters are required to reduce particulate contamination.
Fans on critical fog nodes should be redundant, so that if a single fan fails,
the fog node can continue at full capacity on the remaining fan(s). The
thermal deployment scenario will dictate the enclosure used.
There is a strong correlation to power delivery, performance, and heat
dissipation. This needs to be taken in consideration when designing the
overall solution.
5.6.1.3 Modularity
Most fog nodes will be modular. On smaller designs, modularity could consist
of a motherboard containing the common fixed components, and a few
modular sockets into which configurable components may be installed. Most
modular systems also have to trade-off capabilities for that modularity. For
example, a modular adapter may require the use of a different enclosure to
enable the in-field upgrade, but this modularity can also provide more levels
of serviceability.
Examples of configurable components include:

75
 Faster CPUs
 Different RAM components
 Different storage configurations
 Configurable I/O to support both southbound edge interfaces and
northbound networking interfaces.
o Network interfaces, including varying numbers of wire line and
optical interfaces with configurable physical layer options.
o Southbound interfaces like RS232, Modbus, etc.
 Accelerators including FPGAs, etc.
Moderate sized fog platforms may substitute a backplane for the

motherboard, and support modularity through the installation of boards into
that backplane. This is typically seen in near edge or on-premises fog
platforms. The largest fog platforms will resemble high capacity blade
servers, supporting many modules, including high-end multi-socket CPU
farms, large GPU arrays, petabyte class storage, and potentially thousands
of I/O links.
A good example of these sizes in a deployment scenario are the fog

platforms that required to support machine vision. A training system near
the edge would utilize the larger fog platforms to train a neural network. The
moderate sized fog platforms would take the trained model and use that for
inference or recognizing images dynamically across many different video
streams. A smaller fog platform could be embedded in a camera and utilize
an embedded accelerator to recognize images coming off of a singular
camera feed.
5.6.1.4 Module-Module Interconnect
Modular fog platforms may require interconnection between internal

modules. These interconnections may be connected between a motherboard
and daughterboard, or board-board over a backplane. Hundreds of GB/s
may be required for module-module interconnection. The transport could be
wire, optical, or other means. The connection between modules is
sometimes referred to as a fabric.
In the largest fog platform, one or two fabric modules can be used as a
central hub. The CPU, accelerator, storage and networking modules are used
as spokes in a star topology. Ideally these interconnect facilities should
conform to open standards like PCI Express or Ethernet to facilitate a widely
interoperable hardware ecosystem.
76
5.6.2 Hardware Virtualization and Containers
Hardware-based virtualization mechanisms are available in almost all

processor hardware that would be used to implement fog platforms. It may
also play an important role in system security. Hardware virtualization for
I/O and compute enables multiple entities to share the same physical
system. Virtualization is also very useful in ensuring that virtual machines
(VMs) may not utilize instructions or system components that they are not
by design supposed to utilize.
Containers are a relatively new technology. Containers may offer a lower

weight isolation mechanism within a fog computing environment. The
isolation guarantees are only made by the OS and not fully based in silicon.
This shifts the isolation requirements from the silicon to the software running
on the silicon. The decision to use containers or VMs for isolation are usually
based on security considerations for a given use case. We will discuss
containers in greater depth in the software view.
Software Architecture View
The software view of the OpenFog RA is composed of software running on a

platform that is comprised of one or more node views coupled with other
components to create a system addressing a given scenario. The
stakeholders of the software view include the system integrators, software
architects, solution designers, and application developers of a fog computing
environment. The software running on fog platforms is used to satisfy a
given deployment scenario. A robust fog deployment requires that the
relationship between a fog node, fog platform, and fog software are
seamless.
5.7.1 Software View Layers
As shown in the figure below, the software of the fog node can be separated
into three layers that sit on top of the platform hardware layer.
Application Services: Services that are dependent on infrastructure

provided by the other two layers, fulfill specific end use case requirements,
and solve domain specific needs.
Application Support: Infrastructure software that does not fulfill any use
case on its own, but helps to support and facilitate multiple application
services.
77
Node Management and Software Backplane: General operation and
management of the node and its communications with other nodes and
systems. IB in the diagram refers to In Band management. This is generally
how software interacts with the management subsystem.
Figure 23 Software Architecture View
5.7.1.1 Software Backplane and Node Management
5.7.1.1.1 Software Backplane
The software backplane is required to run any software on the node and
facilitate node-to-node communications (east-west as well as north-south).
This includes:
 OS: May include unikernels that operate on top of a virtualization

layer and extend all the way to application micro services.
 Software drivers and firmware: Interface with and enable
hardware.
 Communication services: Enable communications and might help
to define software-defined networks and protocol stacks
 File system software
 Software virtualization: To provide hardware-based virtualization
support for running software and application micro services.
 Containerization: To provide OS based isolation support for
running software and application micro services.
78
Software containers provide a good mechanism for fine-grained separation
of applications and micro services running on the software backplane. A
software container, unlike a VM, often does not require or contain a separate
OS. A container uses resource isolation of the CPU, memory, block I/O,
network, etc., and separate namespaces to isolate the application's view of
the operating system.
Within a container, applications can be configured, resources isolated, and
services restricted. Multiple containers share the same kernel, but each
container can be constrained to only use a defined amount of resources,
such as CPU, memory, or I/O.
Containers facilitate highly distributed systems by allowing multiple
applications to run on a single physical compute node, across multiple VMs,
and across multiple physical compute nodes. This is a critical capability for
the elastic compute environment needed for fog computing.
Security of the software backplane is the component by which trust is

established in the software layers above the backplane. The backplane
should provide a means of verification of the application support and
services layers by use of the chain of trust established by the node and
platform. This verification may extend to remote attestation to an external
system.
In order for containers, their application services, and micro services to be

securely initialized and provide their intended services, the software
backplane should enable the root of trust to be extended. Because the
software backplane manages the creation and retirement of trusted
execution environments and/or containers at the upper layers of the
software stack, it is critical that the upper layers are able to verify attesting
entities. The software backplane should define a policy by which it enforces
responses based off data it receives from other devices. These policies may
be as specific as only allowing communication with devices that use specific
firewall rules, application mix, and installed patches. Or it may be more open
policies, such as requiring a specific OS release.
The Software Backplane layer orchestrates thing-to-fog, fog-to-fog and fog-

to-cloud communications. This layer must and should provide data
confidentiality and integrity services to protect communications in all
directions and must and should enforce data-origin and peer-to-peer
authentication on connectionless and connection-oriented communications
respectively. Nonrepudiation of data origin and destination may also be
provided in this layer to north, fog-to-fog in order to support remote
attestation required to support trusted computing. These services should
derive their security and trustworthiness from the security credentials issued
79
by the Security Management maintained in the Hardware Root-of-Trust. All
communications (wired or wireless) across the Software Backplane must use
a base list of standard cryptographic functions to provide confidentiality,
integrity, authentication and non-repudiation services[Section 10.1.1]. For
performance reasons, these cryptographic functions are often performed by
the Crypto Accelerators installed in the fog node. The following describes
some additional facets of the Software Backplane:
 Service discovery: Is essential when multiple fog deployments

need to come together they are able to create multiple trust
boundaries in an ad-hoc manner for the purpose of cooperative
information exchange and computation. Establishment of trust
between fog deployments for transient collaboration requires well-
founded trust framework and trust provider service graph.
 Node discovery: Applies to intra-fog discovery in a clustered
deployment. When a new fog system is added to the cluster, it will
broadcast its presence and joins the cluster. From then on, this
node is available for sharing the workload.
 State management: Support for both stateful and stateless
computational models. Stateful computational model can
externalize the state or store the state within the fog cluster
through a resilient replica model. Consensus algorithms may be
utilized to ensure that multiple replicas of the same entity are
eventually synchronized for preventing data loss. Externalized state
requires the help of session/state micro-services that run on top of
well-known database and storage technologies.
 Publication and subscription management: Application layers
running on top of fabric runtime will need infrastructure support for
publication of events, notification of state changes, and broadcast
of messages. The publication and subscription management
mechanism will support temporal as well as standing subscriptions
and pluggable notification endpoints. The runtime stays abstract; it
is the applications layer that composes payload pushes to the
destination endpoints through the runtime layer.
5.7.1.1.2 Node Management (In Band)
When describing management, it is often acceptable to overload the term
“node” with system. The fog In Band management layer is responsible for
keeping the hardware and software of the fog node or system configured to
the desired state, as well as keeping it running at specified levels for
availability, resilience, and performance. The composition of the
management layer will vary, in order to support fog nodes designed with
different capabilities. For embedded and standalone deployment models,
node management may be handled from a remote location. The hardware
80
platform management subsystem of the fog nodes cooperates with software
on the main processor to perform this node management. The following are
the list of capabilities needed by each fog node:
 Configuration management: Operating system and application

support configuration is managed through a software agent that
maintains the desired state of the OS and the application runtime.
Agents are an optional aspect of node management as the deployment
cost.
 Operational management: Operational telemetry of the fog nodes
will be captured, stored, and presented for systems management
personnel and automated systems responsible for monitoring the
infrastructure. The information includes network operational events
and alarms generated by the network, OS, and applications. The
monitoring systems will manage the operational workflows for acting
on critical alarms. The remediation of those alarms can be automated
or manual depending upon the alert.
 Security management: Security management includes key
management, crypto suite management, identity management, and
security policy management.
 Capacity management: Monitor the capacity and page in additional
compute, networking, and storage resources as demanded by the
workload.
 Availability management: Critical infrastructure requires automatic
healing in the event of the malfunction or crash of the software or
hardware. The workload will be relocated to a different hardware node
if hardware fails. In the event of a software failure, the VM or
container may be recycled. Enough reserve system capacity should be
kept in a ready state to meet any SLAs for the given scenario.
5.7.1.2 Application Support

Application support includes a broad spectrum of software used by and often
shared by multiple applications (micro services). Application support is
neither domain nor application specific, but may be dependent on the
underlying layers (including virtualization, hardware, etc.). As shown in the
figure below, depending on the deployment type or application, support
software may be provided in multiple forms (e.g., the use of multiple
application storage databases may be required in some deployments on
some nodes).
81
Figure 24 Application Support
While not required, elements within the application support layer may often
be containerized deployed in some form of virtualization as provided by the
software backplane. As examples, a message broker or NoSQL database
which are used by several applications (within the applications services layer
– see the figure below) within the Fog node, may itself be independently
containerized and offer its supporting capability via the bounds of that
containerization.
Figure 25 Containerization for Application Support
The containerization or virtualization of any support layer capability provides

looser coupling, additional security, and can even allow the backplane to
scale the supporting layer more quickly/easily.
Application support includes the following:

82
 Application management: The provisioning, verification, updating,
and general management of the application support software, as well
as application micro services. The same mechanisms also manage
configuration images for accelerators like FPGAs and GPUs. Application
management covers:
o Application provisioning: The network runtime will host

provisioning agents for receiving and acting on application
provisioning requests from the management system. The
management software will then use the application manifest
to pick the right image from a versioned image store. The
provisioning agents will help with the rollback requests.
o Image (application bits) management: The fabric runtime
layer will need the support of an image management
framework hosted as a part of the fabric control
infrastructure. The image management interface may include
image verification for trust, malware, versioning, and
dependency management.
o Image verification: Fabric runtime, by default, only runs an
image that is verified to be safe. This requires the fabric
controller to support code authentication schemes like PKI.
o Version management: The image management framework
will allow the deployment of a particular version of the image.
The create, read, update, and delete (CRUD) functions for
storing the image are essential for populating the image store
and provisioning a particular version in case of rollbacks.
o Transparent updates: Depending on the stateful or
stateless behavior of the hosted application endpoints, the
fabric runtime will spread multiple instances of the same
application into update zones. This enables the runtime to
visit and update one update zone at a time, ensuring that the
entire application is not taken down.
 Runtime engines: VMs, containers, platform runtimes, program

language libraries, and executables provide the execution
environments for applications and micro services. Examples
include: Java Virtual Machines, Node.js (JavaScript runtime), NET
Framework, Python Standard Library and runtime executables.
 Application servers: Application or web server hosting micro
services or other node supporting infrastructure or applications.
Examples include: Wildfly/JBoss, Tomcat, and Zend Server.
 Messages and events (buses or brokers): Support for message
and event-based applications and micro service communications
(often categorized under message oriented middleware, message
83
broker, message bus, etc.). Examples include: message DDS,
ActiveMQ, and ZeroMQ.
 Security services: Support for application security including
encryption services, identity brokers, etc. Security services within
the application support layer may also include deep packet
inspection, intrusion detection and prevention systems, as well as
system and network event monitoring, content filtering, and
parental control. Details of these are included in the security
appendices.
 Application data management/storage/persistence:
Application data transformation capability and storage to include
durable persistent as well as in-memory caches. Persistent storage
may include both SQL and NoSQL databases, but other forms of
durable storage should be considered, such as in-memory
databases and caches (to address latency and performance
concerns). Examples include: SQLite (SQL), Cassandra & Mongo
(NoSQL), and Redis or Gemfire (in-memory databases).
Considerations within this layer include:
o Encoding/decoding/transcoding:
 Decoding: From binary to JSON for application layer

processing. For example, protocol translation from OPC
UA/DDS/LONWORKS binary to JSON.
 Encoding: From application layer payload to binary for
transmission. For example, from JSON into OPC UA
binary
 Transcoding: Translation of a data structure from one
format to another format within the same layer,
possibly using a gateway
 Encryption/decryption: Data in motion as well as
data at rest
o Information persistence/cache
 Enable storage of structured and unstructured data

 Durable/non-durable (e.g., in-memory, local disk,
external storage service)
 Pluggable into the encoding/decoding/transcoding pipe
and filter process
 Support streaming and batching models
 Support multi-tenancy (e.g., information isolation
between scopes/profiles)
 Store and forward capability
84
 Special support for digital media content (e.g., digital
rights management, add insertion)
 Analytics tools and frameworks: Examples include Spark, Hadoop,

and other MapReduce types of technologies.
5.7.1.3 Application Services Layer

Fog node applications will vary greatly based on deployment, use case, and
resource availability. Fog computing applications are composed of a loosely
coupled collection of micro services. As shown in Figure 31, these services
can be separated into layers based on their roles.
Figure 26 Application Services Layer
Applications, as with application supporting services, may run inside of

containerized/virtualized environments as offered by the software backplane.
These applications may take advantage of support layer services (like a
database or message broker) that are containerized or may run on or be
deployed to containerized supporting services (like a runtime engine).
85
Figure 27 Containerization for Application Support
Fog connector services are used at the "south side" of the architectural
application layer – that is the interfaces in the direction of the IoT things.
This includes legacy transports like Modbus et al. These micro services
contain a set of APIs for enabling higher-layer fog services to communicate
with devices, sensors, actuators, and other platforms using the edge
protocol of choice. Fog connectors operate on top of the protocol abstraction
layer to translate the data produced and communicated by the physical
things into a common data structures/formats and then send that into the
core services.
Core services separate the edge from the enterprise. Core services collect
data from the edge and make it available to other services and systems
above, such as the cloud. Core services often route enterprise and other
system requests to the appropriate edge resource, sometimes translating
the requests for edge devices. This translation is a function of the fog
connectors, which contain a set of APIs for receiving and translating
commands from higher-level fog computing services (or the cloud) to edge
devices for actuation.
Supporting services encompass a wide range of micro services that provide

normal software application duties, such as logging, scheduling, service
registration, and data clean up.
Analytics services may include both reactive as well as predictive capability.

Closer to the edge, fog nodes will likely have services that are more reactive
in nature. Fog nodes with more processing power and capability (usually
farther from the edge) will have more predictive capability based on machine
learning and other cognitive services.
86
Reactive analytics look at the raw incoming data and monitors for change =
outside of expected norms. This includes but is not limited to:
 Critical event processing

 Simple anomaly detection
 Data out-of-bounds alert triggering
 Sensor fusion with stream processing
 Supervisory / distributed control
 State machine and state machine engine
 Expression language (versus SDK) for moving around the state
machine
 SDK and/or API set to provide/update rules to the event processing
engines.
Predictive analytics are defined as forecasting analytics. This includes but is

not limited to:
 Fog node machine learning can support fog-only or hybrid models

where some aspects (say training) are performed in the Cloud, while
more processing intensive and high scale aspects (say an inference
engine) execute in the fog. Models may be generated in the cloud and
communicated down to the fog node agents for use.
 Connectivity to machine learning or other predictive-styled analytics
engines that may be running off-node.
 The development of actionable intelligence gain from a collection of
sensor/devices that typically could not be derived from a single sensor
or device (referred to as sensor fusion). Data can be fused from similar
sensors measuring in parallel (like an array of security cameras), or
different sensors monitoring different (but interrelated) parameters.
Sensor fusion may also include information securely collected from
outside the node, such as information from the Internet.
 SDKs and tools focused on how to connect predictive or machine
learning algorithms to the stream of data, model creation, etc.
Integration services allow outside fog nodes to register for data of interest
collected or generated by the fog note and dictate where, when, how, and in
what format the data should be delivered. For example, a client may request
that all temperature-related data be sent via REST to a prescribed address,
every hour, encrypted, and in JSON format. Integration services then
provide the means to deliver the data using a pipe/filter mechanism as
specified at the time of client registration.
User interface services are micro services dedicated to the display of:
87
 Data collected and managed at the fog node
 Status and operation of services operating at the fog node
 Results of analytics processing at the fog node
 System management and fog node operations
88
6 Adherence to OpenFog Reference
Architecture
The OpenFog Consortium intends to partner with standards development
organizations and provide detailed requirements to facilitate a deeper levels
of interoperability. This will take time, as establishing new standards are a
long and ongoing process. Prior to finalization of these detailed standards,
the Consortium is laying the groundwork for component level interoperability
and eventually certification. The OpenFog Testbed working group, in
conjunction with the Technical Committee, will provide the details for
OpenFog adherence to the architectural principles and various views that will
be shown through our testbed initiatives. They will be used for the various
technologies that can support the OpenFog RA and overall solutions to a
given scenario. A technology that is used to facilitate part of a fog solution is
termed OpenFog Technology Ready.
Figure 28 OpenFog Technology Ready
An OpenFog architectural E2E solution to a scenario is what we call OpenFog

Ready.
Figure 29 OpenFog Ready
Prior to standardization the Consortium will leverage its technical working

groups to score various implementations that claim to OpenFog or fog
computing implementations. This scoring and the subsequent appeals
89
process will be visible to members and will be published to highlight and
recognize progressive fog computing implementations and those that do not
follow the OpenFog architectural principles and OpenFog reference
architecture.
90
7 An End-to-End Deployment Use Case
Earlier, we described how the OpenFog RA would be used by developers,
designers, and architects to create solutions for vertical market use cases. In
the chapters that followed, the OpenFog RA provides the details of a generic
fog platform. In this section, we address an end-to-end use case.
Airport Visual Security

Visual security (surveillance) for airports provides an excellent end-to-end
scenario for fog computing. It illustrates the complex, data-intensive
demands required for real-time information collection, sharing, analysis, and
action.
First, let’s look at the passenger’s journey:
 Leaves from home and drives to the airport
 Parks in the long-term parking garage
 Takes bags to airport security checkpoint
 Bags are scanned and checked in
 Checks in through security and proceeds to boarding gate
 Upon arrival, retrieves bags
 Proceeds to rental car agency; leaves airport
Figure 30 Airport Scenario
This travel scenario is without incident. But when we introduce any type or
number of threats into this scenario, the visual security requirements
become infinitely more complicated. For example:
 The vehicle entering the airport is stolen

 The passenger’s name is on a no-fly list
 The passenger leaves his luggage someplace in the airport
91
 The passenger’s luggage doesn’t arrive with the flight
 The luggage is scanned and loaded on the plane, but it is not picked
up by the correct passenger.
 An imposter steals (or switches) a boarding pass with another
passenger and gets on someone else’s flight.
 The passenger takes someone else’s luggage at the arrival terminal
Catching these possible threats requires an extensive network of surveillance

cameras across the both airports (several thousand cameras at each
airport). An IP H.264 or H.265 camera produces 12Mbps at 30fps (frames
per second) or approximately 1TB/day per camera that must be transmitted
to security personnel. Or, more likely, the video streams will be forwarded to
local machines for scanning and analysis.
In addition, law enforcement will need data originating from multiple

systems about the passenger’s trip, from the point of origination to arrival.
Finally, all of this video and data must be integrated with a real-time threat
assessment and remediation system.
7.1.1 Cloud and Edge Approaches

In an edge-to-cloud design, every camera (edge device) in the airport
transmits directly to the cloud for processing, as well as the other relevant
data collected from the passenger’s travel records.
Advantages Disadvantages
Edge-to-  Store shared data  Latency (inability to process
Cloud in a common images and alert authorities with
Approach location millisecond turnaround)
 Historical  High cost of data transfer
analytics for  Reliance on always available cloud
threat prevention
planning
Edge-only  Low latency  Limitations in sharing data and
Approach information across systems within
the airport.
 Limitations with sharing data
between airports in near real time
In both of these approaches there are advantages and disadvantages.

However, in both cases we believe that the disadvantages in Edge-to-Cloud
and Edge-Only drive the requirement for Fog computing.
92
7.1.2 Fog Computing Approach
The power of fog computing is that we can insert computation where it is

needed to address the given problem. Before going into the solution let’s
look at the OpenFog pillars and how they relate to our specific use case.
Figure 31 Key pillars of the OpenFog Architecture
The OpenFog pillars covered in great detail in chapter 4 are present

throughout the airport visual security E2E architecture. There are some that
require special attention because of this specific deployment scenario.
 Security: The airport visual security scenario is a physically
distributed Fog deployment. Thus, physical possession is in scope for
our security analysis. Transportation and storage of data must also be
secure as much of the data which may contain personally identifiable
information.
93
 Scalability: is essential for OpenFog implementations to adapt with
the business needs as it relates to system cost and performance.
When you add a new airport terminal, gate, or additional sensors and
equipment the solution must scale and not require a completely new
deployment.
 Open: Openness is essential for the success of a ubiquitous fog
computing ecosystem for IoT platforms and applications. Proprietary or
single vendor solutions can result in limited supplier diversity, which
can have a negative impact on system cost, quality and innovation.
 RAS: The various aspects of the solution must be reliable, available,
and serviceable which includes orchestration of existing or new
resources. As new object recognition models are trained for visual
analytics, these inference engine models should be updated on near
edge devices without impacting availability of the solution.
 Programmability: Visual analytics is utilized to facilitate this scenario
and hence programming at the hardware is utilized. In our
implementation, we may utilized accelerators such as FPGAs to
perform inference on images as they are seen in the scenario.
The OpenFog reference architecture includes several layers, perspectives or

cross cutting concerns, and views to enable an OpenFog implementation for
the airport visual security scenario.
Figure 32 OpenFog Architecture Description
In this visual security scenario, we focus on the platforms, data analytics,

performance, and higher-level software infrastructure.
94
 Performance: indicates that we need to include time to adequately
process the information and provide useful latency sensitive
information to enact an action.
 Platforms: indicates we need to have the correct hardware platforms
with the appropriate accelerators and communications infrastructure
needed to have each Fog node communicate with each other to
address the scenario.
 Data Analytics: means that as we process this information at the
individual Fog nodes, those nodes that are nearest the edge enable
higher level intelligence at each level in the hierarchy.
 Software Infrastructure: indicates we can transition the data,
intelligence, and environments across various Fog node deployments
and enable higher-level computation.
The core aspects of a fog node can also be viewed as compute, storage,
network, accelerators and control.
7.1.2.1 Functions of a Fog Node for Visual Security
Figure 33 Node view for Visual Security
Sensors, Actuators and Control

In the end-to-end fog computing implementation, we start with the sensors
at the edge of the network. Previously, we discussed the number of cameras
in an airport surveillance installation, and the volume of data generated by
these many thousands of cameras. In addition to cameras, there is a wealth
of other edge devices collecting data that can be useful in preventing a
threat. These include:
Physical security sensors  Gates, doors, motion detectors,
95
etc.
Safety sensors  Fire, smoke, heat, and bomb

sensors. Note: these independent
systems can either be on the
Ethernet or stationed behind a fog
node (see next section on legacy
protocol bridges).
 Audio sensors can capture voices
Audio sensors and basic audio
and sounds that may be important
analytics
in detecting and assessing
threats; this audio information can
be forwarded for basic audio
analytics and alerts sent to higher
levels in the hierarchy for
processing.
 RFID sensors can be used to
RFID sensors
gather information from
passengers, such as passport
information
These sensors are connected to the fog node via a multitude of interfaces.
These should be standard open interfaces like PCIe, USB, Ethernet, etc. They
should also support open APIs but their implementation may be proprietary.
This is important to ensure openness. After they are connected, they can be
used by higher level software. For example, the RFID reader would be
hooked up to the Fog node by Ethernet or USB. The fog node can them
utilize the data and provide that to higher level entities.
Protocol Abstraction Layer (Legacy Protocol Bridge): A fog computing

solution doesn’t require a “greenfield” deployment; it assumes that there will
be mixture of analog cameras combined with digital cameras. One method to
convert the analog feed is to utilize accelerators (low cost FPGAs provide a
good implementation option). The fog node needs to be able to take a
multitude of sensors and perform sensor fusion, which, in turns, requires
that the fog node have a variety of physical interfaces for converting legacy
analog to digital. These interfaces include coax, USB, RS232, audio, PCIe,
etc., and system interfaces like SPI. If a system can connect directly to an IP
network, the RA utilizes the software backplane to provide sensor fusion for
higher-level software to utilize. The reality is that in many implementations
of open interfaces and protocols, they have slight variations and hence
require a protocol abstraction layer to effectively utilize them.
96
Network: Each IP-connected camera (and many other sensors) is
connected to a fog node using Ethernet. In our scenario cameras represent
the largest population of sensors. We believe that a singular fog node can
aggregate four to eight camera streams with a 1Gb (camera to node) and
10/1Gb (fog-to-fog) interconnect. Fog nodes require a minimum of 16
Ethernet ports to support camera-to-node, node-to-node, and node-to-cloud
communications. First-level fog nodes may also support multiple non-IP
protocols for aggregating both IP and non-IP traffic (e.g., BLE, Z-Wave, and
badge reader information). Time-sensitive networking (TNS)/ deterministic
communication is supported by the network and the fog node to prioritize
alerts across a busy or noisy network.
Accelerators: The visual security scenario offers many places to include

accelerators. As described above, FPGAs may be used to convert the analog
input into a digital format. In this application, the FPGA may be connected to
the node using a traditional PCIe interface. Accelerators also play a role
when performing visual recognition (face, object, etc.). AlexNet, GoogleNet,
TensorFlow, and other neural networks can be used to accelerate a matching
an image with a threat. This is usually called inference scoring based upon a
given trained model. The recommended interface to the node in this
example is the same for the analog-to-digital format (most likely PCIe). This
enables future upgrade of FPGAs as cost and technology changes over time.
Compute: As a fog node is installed, higher-level software needs to

understand the capabilities by which it processes the data generated by all
connected sensors and cameras. Compression can also be added to the
cameras to reduce the general-purpose computation requirements at the
camera. The next level in the hierarchy would require higher processing to
perform video analytics using general purpose compute resources or
accelerators. In our scenario, it is recommended that most of the compute
be performed in the fog node and not necessarily on the camera.
Storage: For a given camera feed, a fog computing design should capture
24 hours of rolling data. This requires ~1TB of localized storage for each
camera feed. The interface to these devices is either SATA or PCIe; the
medium is either flash-based or spinning disk. In one topology, you can see
how a single fog node can act as the storage for approximately eight camera
feeds. This is a traditional Network Video Recorder (NVR) function.
In some implementations, storage will be resident on the camera. However,

we believe it is more optimal for the airport visual security scenario to
configure the storage on the hierarchy of fog nodes. This will mean reduced
camera cost and combining NVR functions with the video analytics
97
capabilities in the same fog node. However, this will increase the compute
capability requirements at the next level in the hierarchy.
Management (OOB): If the fog node has updateable firmware and

software, the design can specify remote updates. This is part of the
serviceability requirements of the RAS pillar. The RA also requires that the
node implement a mechanism by which a management entity can ascertain
the health of the overall node (minimally, a healthy or failed state).
Traditionally, these notifications are accessible via I2C or SMBus interfaces
and routed to a management controller. This function can also be made
available to an in-band interface to the operating system. This allows higher-
level software to interface with the management subsystem. The RA
recommends that any component that can be field repaired or removed
should have a health indication which can be accessed via this interface.
Security: Nodes must have an immutable hardware-based root of trust for

verification. This is to ensure that agents capable of communicating in the
airport are operating from good firmware. The RA also recommends a
hardware-based root of trust for measurement to provide attestation for the
software and firmware on the node. This can be supported by a TPM or
firmware TPM providing it does not impact the overall hardware-based root
of trust.
7.1.2.2 System View for Visual Security
This section describes how the components of the fog system view interact
to support this end-to-end use case.
System View Hardware Virtualization: The hardware of the hierarchy of

fog nodes used in the visual security example should support virtualization
techniques. Applications should also avoid being dependent upon which level
98
of the hierarchy their various processes are hosted on. If one fog node in the
network hierarchy is overloaded or down, the system level orchestration will
move its responsibilities to adjacent nodes in the same level or adjacent
levels. Processing, accelerators, storage and networking functions should all
be virtualized to maximize the efficiency and flexibility of the fog system.
Virtualization may also incorporate aspects of containerization, depending
upon the software layers that run on the hardware.
System View Management: The sophistication of the management system

must be balanced with the simplicity of usage. The RA addresses a network-
level, end-to-end view of installation, configuration, operations, monitoring,
troubleshooting, repair, growth and decommissioning of all elements of the
system. System-level node management must be as autonomous as possible
to minimize complexity of nodes that a given fog infrastructure can handle.
By doing this we can increase the overall scalability of the solution.
Additional management includes a higher-level hardware manager that is

responsible for all system management. The management system monitors
the health of all cameras, storage and other hardware assets. This
management interfaces with the software backplane to satisfy the overall
systems management.
System View Security: At the system level, all the fog nodes in the
hierarchy must cooperate to ensure the network remains secure. This
includes:
 Nodes in higher levels of the hierarchy should monitor the functions of

the lower level nodes to ensure no ongoing or emergent security threats
exist.
 Peer-level nodes on the same hierarchy level should monitor their
neighbors to detect security threats.
 All node-to-thing and node-to-node communications links should be
encrypted and monitored for suspicious traffic.
 Physical security, including tampering, must also be monitored.
 If a system is tampered with, the management subsystem must be
notified so that appropriate security measures can be executed.
 Communication pathways between nodes must be encrypted.
 Visual security scenarios also include capture of personally identifiable
information (PII), so any data captures must be encrypted while at rest.
This includes end-to-end network security.
System View Network: The network links that interconnect the nodes in
the visual security architecture transport multiple types of traffic.
99
 At the lowest level is the interconnection of adjacent near-edge fog
nodes. These are typically direct point-to-point Ethernet links,
operating at 1 or 10Gbs. They link together the composite picture of
the fused sensor readings across all nodes. For example, stitching
together all camera images on a given fog node with a subset of the
images from selected cameras on adjacent fog nodes.
 Another type of interconnect is between each near-edge fog node and
the second level fog node that serves it. These links usually carry
higher level messages and analytics data that has been distilled by the
near-edge fog nodes.
 Another interconnect type is between the second and third level fog
nodes. The third level fog nodes are master controllers for the entire
visual security implementation. Redundant links are provided for load
balancing and fault tolerance. These links can carry significant traffic
bandwidth, and run for kilometer lengths (usually over fiber). They are
typically 10 or 100GE IP connections.
 Finally, there a set of links between the third-level fog nodes and the
cloud backbone.
System View Accelerators: The airport visual security scenario studied

here may include hundreds and thousands of cameras. Each of which will
produce images that must be carefully analyzed in real-time to detect a
large number of conditions. The accuracy of the analytics for visual security
is vital, to avoid swamping the first responders with false alarms, or worse,
missing some threat scenario that should generate an alarm. Computation
accelerators make these intense computations fast and energy efficient.
Some combination of accelerators (such as FPGAs) at the near-edge fog
nodes, and even larger farms of accelerators at the second and third level of
the fog hierarchy should be used in the design. Accelerators at the third level
may be used in image training (facial recognition), while accelerators in first
level fog nodes would be used for inference on a trained model.
System View Compute: General-purpose computation is vital at all levels

of the system view. These compute resources will typically be multi-core
processors, sometimes configured as multi-socket servers at higher levels in
the hierarchy. Significant memory (from tens to hundreds of GB) are
required on each fog node to avoid performance bottlenecks in video-
intensive applications. System-level compute resources are required for
things like:
 Executing tasks (such as control algorithms and user interfaces) that

can’t be more efficiently run on accelerators or require high single
thread performance
 Managing the networking capabilities of the fog system
100
 Incorporating license plate recognition scenarios.
 The following are usually offloaded to accelerators in our example, but
are kept here for reference:
o Facial recognition.
o People counting
o Threat detection
System View Storage: Storage will be required at all levels of the fog
hierarchy. During the normal operation of the system, some visual data will
be transmitted from the lower levels of the hierarchy to upper levels and
eventually to the cloud.
The storage on the near edge fog nodes should be sized to keep
approximately 24 hours of full-resolution video online locally. For a typical
fog node with about eight 4K resolution cameras, assuming a compressed
data rate of 10Mb/s per camera, this requires just under one terabyte of
storage. Note: data from the non-camera sensors on the fog nodes does not
contribute significantly to the storage requirement. This storage may be
implemented in many different ways (there are NVR solutions that address
this capture requirement). However, storage is finite and the RA balances
the need to retain older data while addressing near term concerns. In these
cases, we believe it is acceptable to have a mechanism to reclaim older
storage for new purposes providing the user explicitly opts in for this
behavior.
Before old data is overwritten on each near edge fog node, the
recommendation is to convert it to a lower resolution (e.g., 720P or 480P)
and forwarding the data to the second level fog nodes. The second level fog
nodes accept these down-sampled streams from all the near-edge fog nodes
they support, and store it for 30 days (in this scenario). This may require
approximately 20TB of storage (assuming down-sampling of all video
streams to 1Mb/s), requiring the second level fog nodes to have significantly
larger storage arrays, perhaps using rotating disks and RAID array
techniques. The third-level fog nodes are basically big servers, and will
probably require big data style storage strategies. The benefits of keeping
this older data in the fog is that we can reduce cloud transmission costs. If
we do decide to transmit the down-sampled data, the cost of transport is
less than if we had to send up the 4K or 1080P image. As described earlier,
“data at rest” on the storage device should be encrypted to protect against
physical theft and compromise of PII (Personally Identifiable Information).
System View Hardware Platform Infrastructure: The hardware platform

infrastructure has a common set of infrastructure components—including
chassis, backplane, power, cooling operating system and management—into
101
which hardware and software modules are plugged to customize the function
of specific fog nodes. The same infrastructure elements can then be used to
produce lightly populated nodes without significant computational,
networking or storage requirements. The same elements could also be used
in fully-equipped fog nodes (perhaps higher in the hierarchy supporting
greater application demands).
System View Protocol Abstraction Layer: Protocol abstraction permits

the fog network to operate independently of the specific protocols used in
sensors and actuators or in the cloud. Adaptation logic (both hardware and
software) converts native protocols used outside the fog network to internal
standard protocols, storage formats and data abstraction models within the
fog hierarchy. This abstraction is one of the reasons that fog computing can
easily support the diversity required for this end-to-end airport visual
security scenario.
System View Sensors, Actuators and Control: As described in the

section on sensors, actuators and control above, there is a very rich set of
sensors and actuators involved in the airport visual security scenario. Using
sensor fusion techniques, the fog hierarchy will assemble the inputs from
many sensors into a cohesive view of a threat situation at the airport.
Equally important, fog computing provides the low latency required for swift
response. Latency is also described under Performance.
System View Performance: Low latency is a key attribute of fog

computing, as delayed detection and analysis of a threat is unacceptable. To
achieve this low latency, the analytics algorithms have significant
performance requirements, related to factors like accuracy, object database
size, CPU utilization, energy efficiency, etc.
System View Scale: The airport depicted in our example is of moderate

size, with 24 gates. A fog network will scale to support changing airport
requirements. For example:
 The same fog network design can be scaled for small airports to hub
airports.
 New algorithms will be introduced continually, requiring additional
compute, accelerator, network and storage capabilities; the fog
architecture is designed so that most modules can be upgraded
without requiring the complete replacement of the infrastructure.
 Using the Fog as a Service (FaaS) model, the fog hierarchy will scale
to support many different tenants with widely different needs, all on a
single hierarchical network owned by a single landlord (e.g., airport
authority).
102
7.1.2.3 Software Infrastructure View for Visual Security
As shown in figure below, the software of a fog node can be separated into
three layers. The functions that facilitate the general operation and
management of the node and its communications with other nodes/systems
are found in the backplane and node management layers.
Infrastructure software that does not fulfill any use case on its own, but
helps to support and facilitate multiple application services is found in the
application support layer.
Services that are dependent on infrastructure provided by the other two

layers fulfill specific use case requirements—such as this visual security
scenario—and solve domain specific needs are provided in the application
services layer.
7.1.3 Application to Airport Visual Security

Now that we have analyzed different aspects of the architecture that should
be applied to our solution we will further investigate the requirements and
assumptions.
The following picture provides a simple view of an airport terminal for our
scenario. There are entrances, parking structures, security stations, etc. We
103
will refer to the diagram for various usages like License Plate Recognition as
vehicles enter the airport property.
Figure 34 OpenFog Approach to Visual Security Scenario
We will also deploy multiple fog nodes in each airport and at different levels
in the hierarchy. There may also be a fog node that is responsible for the
entire airport and ensuring that interoperability across systems to achieve
the visual security mandate is in upheld. This is also important so that
airports can share normalized information. Additionally, each fog node may
be connected to another level in the hierarchy. These fog nodes work in
concert to satisfy the requirements of the scenario.
We have several areas to address which we include but are not limited to:
1. License Plate Recognition as vehicles enter the airport property.
2. Passenger Arrival/Departure
a. Parking structures where passengers may exit vehicles and walk
into the airport facility.
b. Arrivals is also a location where passengers may walk into the
airport facility.
3. Passenger Security screening where the passengers are required to
provide identification and boarding passes.
4. Terminals where screened passengers may walk to their gate, shop,
and eventually leave the airport.
104
When the passengers leave the airport, their information should be made
available to the airport where the plane is landing. This is represented in the
following figure with the fog node at a higher level in the hierarchy.
Figure 35 OpenFog realized for Visual Security
The following is a description of the various physical fog nodes that would be
deployed in our scenario.
 Fog nodes for License Plate Recognition (LPR):
o Fog nodes that are surrounding the airport property. In our
example, these are cameras and security devices. These nodes
will be thin in nature and report in adjacent fog nodes. We
estimate that we can service 4 video streams with a singular fog
node.
 Fog nodes that are stationed around the parking structure and arrival
station:
o These nodes will be comprised of video cameras, and as in the
LPR case and those cameras are connected to fog nodes for
visual analytics.
 Fog nodes that are in the arrivals and departure area (just prior to
security screening). These are the same as the parking structure.
 Fog nodes that support the screening process
o These fog nodes are connected to both passive RFID readers and
other sensors as well as cameras.
 Fog nodes that are in the terminal
o These nodes are connected to cameras and additional sensors.
 Fog nodes that watch ingress and egress of passengers to planes.
o These nodes are connected to cameras and additional sensors.
 Hierarchical fog nodes that support and monitor a grouping of fog
nodes.
105
o These fog nodes work on pre-processed data and perform higher
level functions that support the overall mission of safety and
security of the airport.
A complete representation of the realized solution will include the

interconnectedness, security, and software that runs on all the nodes.
7.1.3.1 Machine Vision for Visual Security
Fog computing dictates that we process the image at the appropriate level in
the hierarchy versus sending it to the cloud for analysis. A good mechanism
to address machine vision requirements for LPR, passenger tracking, people
counting and other usages in this specific scenario is the use of a
Convolutional Neural Network (CNN).
When dealing with CNNs it is common to discuss both training systems and
classification systems. Training systems are used to build a CNN network
topology and compute weights against which the image classifications are
validated. This iteration process of adjusting weights or fine tuning weights
are continued till we achieve satisfactory level of accuracy in classifying.
Once we get to this satisfactory level of accuracy (Say a minimum of 98%
accuracy), we push both the both the topology and the corresponding
weights to Target or classification system (aka inference or scoring). AlexNet
which is a well-known CNN for Object recognition and uses a 1.2 million
training image set to build a 1000 different classifications. This gives an
estimation of images required for given number of classifications. Higher
number of images is always better for training.
Figure 36 Training and Classification system for Machine Vision
106
In the LPR case as vehicles are going through the various lanes there will be
a LPR camera and illumination so that we can capture the license plate
image and send that information to the fog node. There are multiple options
for this part. The camera can just capture and send the picture of the vehicle
and license plate doing more work at the edge of the network or the entire
video stream can be compressed and sent to the fog node for additional
analytics.
Figure 37 Airport License Plate Capture
The LPR fog nodes will be trained on license plate images and once the
license plate is capture the fog node will perform 1) Localization 2) Character
segmentation and 3) Optical Character Recognition (OCR) to determine the
license plate state.
Passenger recognition follows a similar flow of using a training system to

train a model based upon an image database. The models for passenger and
vehicles should also be updated frequently based upon new unclassified
images so that the system will learn over time and increase overall accuracy.
The following diagram show several assumptions in how we will get various
information shared between different agencies and private entities. While
this is a simplification of a real scenario it provides the base requirement of
an optimal system where we can securely share information.
107
Figure 38 Integrated Air Travel Infrastructure - Assumptions
7.1.3.2 Airport Passenger

To help simplify the flow we will start with a standard case of a passenger
utilizing the airport for transportation we showed earlier.
The passenger leaves from home and drives to the airport.
There is a fog node associated with each group of cameras that monitors
vehicles as they come onto the airport. This node is responsible for capturing
the license plate image, performing video analytics, and capturing the face
of the driver as they enter the airport property. Fog nodes should also have
the ability interface directly with RFID readers and other data acquisition
devices and sensors to provide local, high performance identification of
people and objects in the vehicle.
 Data security and privacy concerns are addressed throughout the

network. Camera firmware will be protected by a hardware-based root
of trust for verification and, optionally, measurement. This ensures
that processed images are coming from hardware that is operating in a
known configuration and has not been tampered with.
 Privacy concerns for visual images stored on the camera requires that
all saved data is protected from inspection even with physical
possession. This protection involves strong cryptography on all data
links and storage repositories used to hold images of people. It should
be encrypted when at rest and in transport.
108
 If a stolen or suspicious car is detected as it enters airport property
authorities should be notified to address the situation.
 Images that cannot be accurately detected should be saved and used
for retraining.
Passenger parks in the long-term parking garage
At the garage gate entrance where the passenger collects his parking garage
ticket, the cameras catch the first images and persistent data objects in this
end-to-end scenario. The following figure illustrates the software applications
and edge devices that are collecting useful information for this threat
scenario.
Figure 39 Garage Gate Entrance
Fog node hardware and software performs video analytics on the camera
feeds. This involves an image processing pipeline that can be split across
multiple layers of fog nodes in the hierarchy, or multiple peer nodes to
balance load. The fog network processes the images, and recognizes certain
people or objects. For example, if a license plate is detected that falls on a
“bad vehicle” list, the system can discover this with less than a second
latency, and use that information to control traffic gates (without significant
delay). Similarly, the fog network can perform facial recognition, and detect
people on the no-fly list. Fog is valuable in this case, because the latency is
low (permitting automatic actuation of turnstiles, etc.), and the local
processing and storage of the fog nodes can protect passenger privacy.
109
Objects (like abandoned luggage, for example) can be detected and reacted
to quickly and securely by the fog network.
Figure 40 Central System Analytics
The image processing pipeline is an example of fog’s distributed analytics

capabilities, processing the raw images from the cameras in several steps,
shown in the figure above, and described in detail below.
• Data Filter: Cleans and filters incoming data from a variety of fog
nodes capturing raw sensor data.
• Anomaly Detection (machine learning): Detects different types of
anomaly and asynchronously generates models to provide to the
risk scoring system, for example the detection of a person on the
no-fly list, or abandoned luggage
• Critical Event Processor: Rules engine that monitors incoming
data flags events of importance (based upon airport policies
stored in the fog network) and passes them to the risk scoring
system.
• Risk Scoring System: Generates a risk score for vehicles,
passengers, baggage or other entities known to the system.
Passes high-risk targets to decision support systems.
• Decision Support System: Receives high-risk targets from the
risk scoring system; takes actions automatically or raises alerts.
• Operating actuators (for example, running parking lot gates,
turnstiles, alarms, etc.).
110
• Training System for edge algorithms as described in the earlier
section on machine vision.
Entering Arrivals on the way to the Screening Area
At the airport entrance the security system will have to have multiple
cameras to cover all of the vehicles pulling up and people getting out of cars,
grabbing bags and entering the airport. This can also be after the passenger
parks his car and proceeds to the screening area. A key attribute of this
phase of the process is the tracking of the passenger, and everything he
brought to the airport from his vehicle, through the airport departure area
and to baggage check. Notice how the local fog nodes perform sensor fusion
and data checking/correlation to efficiently implement this step.
Figure 41 Terminal Entrance
The software components for the entrance have many overlaps with the
parking garage software. However, there are many more instances of the
capture components as there are many more cameras. Consequently, fog
node processing is much higher here.
Also, since we are seeing new vehicles we will re-use the Bad Vehicle
System. As the passenger proceeds from the garage to the entrance, the
facial capture and baggage capture components will have received updates
of his face and baggage data objects through their subscription through the
software backplane and related data sharing software services. Multiple
instances of each of these software components will be supported by the
111
software backplane. The agility of fog systems make this sort of re-
use/multiuse possible.
Sophisticated analytics systems process all the information gathered by the
fog hierarchy, normalize it, and send it to the next higher level of the
hierarchy for further processing. These analytics algorithms may employ
machine learning techniques to continuously adapt to ever-changing
conditions and threat models. Each level of the hierarchy will combine the
inputs of cameras and other sensors into a more holistic view of the airport,
further digest the view into a more “concentrated” form, which is then sent
to higher levels, where ultimately airport security policies can be applied,
and any necessary enforcement actions (like denying passage of a vehicle or
person through a barrier) will be carried out. This is how processed data
becomes wisdom
After the visual analytics is performed at the node, it can cross check the
information learned about the driver, car status, and flight status if
applicable and package this information for more detailed processing at the
next level. From there, the fog network can determine if the issue requires
security personnel to be alerted.
Passenger takes bags to airport security checkpoint
Throughout the time when the passenger entered the airport property, a
database should be created with all of the various information obtained. This
includes LPR information, car, and images of the passenger and associated
people. The fog network appends the previously processed images and
database entry with new surveillance images of the luggage area, facial
recognition, and other data related to this passenger, such as bag scans,
updated ticket information, etc. Using this additional information, the fog
network analytics can predict where the passenger should be going,
associate any time he is recognized where his bag is not visible, and/or
associate other risks with his behavior, appearance, etc. This correlation of
multiple security camera images with the output of various other sensors
and the database of information about the passenger is an example of fog’s
advanced sensor fusion capability.
112
Figure 42 Baggage flow
• Using techniques similar to the operations described for the

parking garage, the above figure shows some of the software
components of the fog application that manages the baggage
check-in process.
• Vehicle Capture: converts camera images to license info,
make/model info, parking spot & pair this information with
unique identifier for the vehicle that came through the garage
gate. Provides API for other systems to request raw image based
on id.
• Facial capture: converts camera images into unique person
identifier. Provides API for other systems to request raw image
based on the id.
• Baggage capture: converts camera images into unique baggage
based on the id.
• Data Fusion: Associates vehicle identifier to parking spot.
Associates persons to vehicle identifier. Associates baggage
identifier to person identifier.
• Checker: attempts to match facial capture data against bad
passenger images
• Alerter: responsible for signaling possible issue detected to
centralized tracking system
• Bad Passenger System: registry of persons under surveillance by
authorities and/or people on no-fly lists, on arrest warrants, etc.
113
Data for this system is held locally and updated on a regular
basis.
Bags are scanned and checked in

Cameras will pick up the passenger’s new location. The baggage information
will be updated and added to his database record entry. This data will be
published to the various software components in other parts of the system
for risk analysis.
• The figure shows the baggage screening process. Multiple local

fog nodes support each security line, their number is scalable
depending upon compute needs. A mix of sensors including
cameras, millimeter wave machines, and bomb sensors (sniffers)
outputs are combined with all the context from previous stages
of the process to provide a very effective screening process.
Here are some of the Components used in this step.

• Bomb Sensor: Gathers data from bomb sniffer and passes any
events to the Data Fusion system for correlation and eventually
action
• Facial Capture: Converts camera images into unique person
based on the id.
• Mm Wave Screen: Gathers data from mm Wave Machines, scans
images for problems, alerts airport screeners for additional
screening, and passes any events to the Data Fusion system for
correlation and eventually action.
• Behavior Monitor: Uses various camera feeds to monitor for bad
behavior among people waiting in line. Any events are passed to
the Data Fusion system for correlation with facial and passenger
data.
• Baggage Capture: Converts camera images into unique baggage
identifier. Provides the APIs for other systems to request raw
images based on the identification.
• Data Fusion: Associates passengers in line with baggage,
behavior alerts, MM Wave screen alerts, and proximity to bomb
sniffer alerts.
• Checker: Attempts to match facial capture data against bad
passenger images, and forwards alerts from the sensor systems.
• Alerter: Responsible for signaling possible issues to centralized
tracking system.
• Bad Passenger System: Registry of persons under surveillance
by authorities and/or people on no-fly lists, on arrest warrants,
114
etc. Data for this system is held locally and updated on a regular
basis.
Check in through security and proceed to boarding gate

At this stage the data analytics should have enough information produce
cognition from the various fog nodes – that is the raw data from the
cameras and sensors has been converted by the fog network through the
steps of becoming information, and is now knowledge of the fact that the
passenger has maintained control of his luggage, and it has been analyzed
by the security screening policies. The fog network will now be able to
autonomously determine if there is a credible threat involving the passenger.
This entire process will take milliseconds and no longer than a few seconds.
Namely the gate operator must be notified to make a determination to allow
the passenger to enter the plane, validate that their bags were in his
possession at all times, and that they are allowed to fly.
The system needs to alert the pilot, or in future scenarios the autonomous
plan to not take off. This will be solely based upon the entire suite of
analysis performed in the fog. If the passenger is determined to pose a
minimal threat, all barriers in his path will open with sub-second latency,
and he can board his plane without delay. If he is determined to pose a
threat, several levels of escalation are possible. The system could alert
airport authorities (either in central security control, or by locating and
informing the officer physically nearest to Bob). Barriers he may pass
through like turnstiles or mantraps could hold him. Many other fully
automated or semi-automatic responses are possible, depending upon the
treat level detected by the fog, and airport policy.
115
Figure 43 Departure Gate
o The above figure describes some of the steps the fog network
will take as Bob makes his journey to the plane.
System Components
o Facial Capture: Converts camera images into a unique person
identifier. Provides APIs for other systems to request raw image
based on the identification.
o Behavior Monitor: Uses various camera feeds to monitor for bad
behavior. Any actions that raise flags are passed to the Data
Fusion for correlation with facial and passenger data.
o Baggage Capture: Converts camera images into unique baggage
identifier. Provides APIs for other systems to request raw image
o Tarmac Capture: Uses various camera feeds to monitor the
aircraft and tarmac for unusual behavior, security breaches, and
potential aircraft damage.
o Data Fusion: Associates passengers in gate area (by facial
recognition) with baggage and behavior alerts.
o Airline Passenger Manifest System: Provides data about
passengers that are checked onboard the plane.
o Airline Passenger Baggage System: Provides data about
passenger checked baggage.
o Checker: Final assembly and correlation of data from all
available system sources.
116
 Passenger appears to match credentials provided and face
consistently matches with images captured since entering
the system.
 Passenger baggage is accounted for and consistently
matches baggage captured since entering the system.
 Aircraft does not appear to have been tampered with.
 No warnings or issues have been detected by other
systems since the passenger entered the airport space.
o Export: Responsible for sending all relevant passenger data to
the destination Central Tracking / Action System
o Alerter: Responsible for signaling possible issue detected to
Upon arrival, retrieve bags
Security cameras at the arrival airport have data about the passenger. As
early as the arrival gate, the fog computing network can determine if the
passenger arrived and has retrieved his baggage.
Figure 44 Arrival Gate
• The reverse of the arrival process is followed when the passenger

reaches his destination, and many of the same fog-based processes
will insure safety during this part of the journey, as shown in the
above figure.
The process components include:
117
• Facial Capture: Converts camera images into unique person
• Behavior Monitor: Uses various camera feeds to monitor for
anomalous behavior. Any events are passed to the Data Fusion
system for correlation with facial and passenger data.
• Baggage Capture: Converts camera images into unique baggage
identifiers. Provides APIs for other systems to request raw image
• Data Fusion: Associates passengers in gate area (by facial
recognition) with baggage and behavior alerts.
• Checker: Correlates data from facial capture and baggage
capture final with incoming (imported) passenger data received
from the origin airport. Ensures all passengers originally on the
aircraft exit the aircraft and all expected baggage also exits the
aircraft.
• Import: Responsible for receiving all relevant passenger data
from the origin central tracking / action system into the
destination central tracking / action system
• Alerter: Responsible for signaling possible issue detected to
Note that throughout this process, many fog nodes within an airport, and fog
networks at two different airports must maintain high performance, highly
secure fog node-fog node communications. This is possible because of the
highly secure fog infrastructure on all nodes, and the strong cryptography
applied on all node-node traffic.
Proceed to rental car agency; leaves airport (if authorized)
The data collected required interoperability such that each node can operate
upon and gain higher-level insights to protect. In many cases this data
obtained throughout passenger’s journey can be shared with local
governmental agencies to also track if they can only be in the country for a
certain amount of time, if he is on parole, etc. Assuming he is not a threat in
his destination, they are cleared to rent a car, and the rental car company
can have much higher levels of confidence that the passenger is who he says
he is, and poses minimal danger, because this information is selectively
shared between the arriving airport’s fog systems and its car rental
agencies.
118
 The above scenario about a passenger’s journey serves to illustrate
some of the key attributes of fog. Fog’s distributed processing
capabilities and hierarchy support the sophisticated analytics and sensor
fusion algorithms that analyze their appearance and actions. Fog’s low
latency permits nearly instantaneous reaction to his actions (for
example, opening a barrier in milliseconds, where cloud-based
processing of the same sophistication could take seconds). The highly
secure nature of the OpenFog implementation insures that the
passenger’s privacy is maintained and visibility up to higher levels in the
system hierarchy are constrained. Fog’s reliability insures the system
will continue operating even if a fog node, inter-node link or the
connection to the cloud goes down. Fog’s bandwidth efficiency insures
high bandwidth traffic like video traverses only the most capable links.
This detailed use case as applied to airport visual security scenario is

intended to illustrate the key benefits of the OpenFog Reference
Architecture. It is intended to be used as a reference for those exploring
the application of fog computing to similar concepts and techniques to
solve similar challenging problems.
119
8 Additional Opportunities
The OpenFog Consortium collaborates tightly between its academia/research
and traditional industry members. This allows the Consortium to leverage
the research and publishing focus of academics with the business
requirements of industry. Among the additional areas of fog computing
research are:
 Interactions between fog and cloud computing including dynamic and

secure shifting and sharing of resources.
 Security refinements not covered by existing efforts of industry
associations.
 Enhancements required for deepened management and orchestration
of fog computing.
 Fog based training to support deep learning and machine learning
without requiring cloud.
 Fully development the Fog as a Service (FaaS) model.
 Performance modeling and measurement to ensure that designers and
architects are achieving the proper QoS for a given implementation
scenario.
 Generating rigorous, enumerated requirements to help facilitate higher
levels of interoperability.
 Government and societal impacts of a software-defined autonomous
world of fog computing.
 Environmental impacts of a more optimized computing environment.
 Education, research and development for new engineers and scientists
who will be instrumental in shaping the implementation of the fog
architecture. This will include software development for fog computing.
120
9 Summary and Next Steps
The OpenFog Reference Architecture (OpenFog RA) is the baseline document
in developing an open, interoperable architecture for fog computing. It is the
first step in creating new industry standards to enable interoperability in IoT,
5G, Artificial Intelligence, Tactile Internet, Virtual Reality and other complex
data and network intensive applications.
The OpenFog RA represents an industry commitment toward cooperative,

open and inter-operative fog systems to accelerate advanced deployments in
smart cities, smart energy, smart transportation, smart healthcare and
smart manufacturing. Its eight pillars describe requirements to every part of
the fog supply chain: component manufacturers, system vendors, software
providers, application developers. The OpenFog Consortium believes that
without this open architecture, there will be limited interoperability,
reliability and security, resulting in slower adoption and limited functionality.
The OpenFog Reference Architecture is the first step in creating industry

standards for fog computing. The OpenFog Consortium will establish detailed
guidance, interface with standards organizations such as IEEE on
recommended standards and specify APIs for key interfaces in the reference
architecture over the next year. Our technical community is working on a
suite of follow-on specifications, testbeds which prove the architecture, and
new use cases to enable component-level interoperability. Eventually, this
work will lead to certification of industry elements and systems, based on
compliance to the OpenFog Reference Architecture.
For more information on the work of the OpenFog Consortium, please visit
www.OpenFogConsortium.org.
121
10 Appendix – Deeper Security Analysis
This appendix currently contains an initial discussion of several security
aspects in an OpenFog computing environment. This important discussion
was placed here for a couple of reasons. First, security is perhaps the largest
technical concern among critical IoT systems; hence, we wanted to discuss it
in a special section of the document. Second, the discussion contains highly
specialized technical details, which, if included in the main body of the
document, may impact its readability. Thus, we decided to collect the
materials that are most interested to the security professionals in one place.
Future versions of the Reference Architecture will also include appendices

that describe other high-priority cross-cutting perspectives from the archi-
tectural description including performance, manageability, data analytics and
control with similar levels of detail.
Security Aspects
Security is a critical concern for fog computing. We strongly believe there

must be a common security baseline to ensure basic interoperability and
protection. We are also aware that there exists a combination of regional
and governmental requirements that fog computing must satisfy. The
following sections describe our preliminary attempt to accommodate
diversity in approaches while trying to establish a unified practice in the
security realm of OpenFog Architecture.
10.1.1 Cryptographic Functions
Cryptography provides mechanisms to implement security services such as

confidentiality, integrity, authentication and non-repudiation. Cryptographic
functions can be implemented in a Platform Security Processor (PSP) to
protect cryptographic keys and security policies, which then protect other
objects1. Cryptographic functions can also be used to provide a secure
execution environment for trusted software, and protect its memory,
1 The cryptographic functions used by software executing on OpenFog

platform may not necessarily be implemented in the Platform Security
Processor (PSP); they can be implemented independently in software or
hardware outside of the PSP.
122
storage, and communications.
The current version of the document describes the initial base list of required
standard cryptographic algorithms that MUST be available on all OpenFog
nodes. Requiring this minimum set of algorithms is intended to guarantee
interoperability among OpenFog nodes. We understand that this initial list is
limited in its ability to enable global interoperability. Going forward OpenFog
is making it a priority to develop a more complete list that includes the set
of standard algorithms of regional standards bodies from, for example,
Europe, China, Japan, and the US.
There are three basic types of cryptographic functions:

 Symmetric (or Secret-Key) Ciphers for confidentiality protection;
 Cryptographic Hash Functions for integrity protection and
authentication of communicating parties2
 Asymmetric (or Public-Key) Ciphers for generating secret keys,
establishing long-term security credentials and providing non-
repudiation services.
The NIST FIPS 140-2 specification [ref-a] defines the security requirements
for cryptographic modules. This specification covers a list of approved
cryptographic functions as well as a formal process for validating the
implementation of these functions in conformance to the specification. The
OpenFog Reference Architecture adopts a subset of FIPS 140-2 approved
cryptographic functions as described below in order to guarantees a base
level of interoperability among its components. The formal validation of the
cryptographic module (i.e. FIPS 140-2 certification) is left as an option for
each vendor. Due to the growing importance of FIPS 140-2, vendors are
encouraged to subject their products to FIPS 140-2 certification.
The OpenFog cryptographic module MUST support the following FIPS

approved cryptographic functions at a minimum:
 Symmetric Key Ciphers3
o AES (with at least 128-bit keys)
o Triple-DES
 Asymmetric Key Ciphers
2 Message Authentication Codes (MAC) can be used for authentication, but

not for non-repudiation.
3 The Escrowed Encryption Standard (ESS) was withdrawn on December 31,
2015.
123
o ℤ𝑝 , ℤ∗𝑛 Based: DH, RSA, DSA
o Elliptic Curve Based: ECDH, ECDSA, ECQV4
 Cryptographic Hash Functions5
o SHA-224, SHA-256, SHA-384, SHA-512, SHA-512/224, SHA-
512/256
 Random Number Generators
o See Annex C:
Approved Random Number Generators for FIPS PUB 140-2,
Security Requirements for Cryptographic Modules
 Message Authentication Codes
o CCM, GCM, GMAC6, CMAC, HMAC7
As defined in Annex A of FIPS 140-2 [ref-b], due to the successful breaking

of cryptographic algorithms in use or the availability of more powerful
computing techniques, NIST provides guidance on updating the list of
approved cryptographic functions. Please refer to NIST “Recommendation for
Transitioning the Use of Cryptographic Algorithms and Key Lengths” [ref-c]
for the latest guidance. Subsequent versions of this document will consider a
transition approach that works for all regional cryptographic algorithms.
Note: Compliance is not security; some of the FIPS approved cryptographic

functions may be considered weaker in strength and leave their
implementation open to potential compromise. In the design of various
OpenFog components, the cryptographic functions selected for their
implementation SHOULD be appropriate for their use and in agreement with
the findings from their stakeholder’s threat assessment.
10.1.1.1 Crypto Accelerators
Cryptographic functions can be implemented either in software or in a

hardware accelerator. While such a hardware accelerator provides an
important security function for the system, the device itself must also be
secure. If it is implemented in a virtual environment, it must be
implemented as a hardware virtualized device so that it can be securely
accessed by multiple independent processes and/or VMs while maintaining a
4 ECQV is used for Implicit Certification.

5 SHA-1 has been deprecated since 2010.
6 See NIST 800-38D.
7 See RFC 2104 and FIPS 198.
124
context specific to each virtual instance and providing access protection for
each address space that “owns” the virtual interface.
Many platforms implement a hardware security processor of some kind (e.g.,

TPM, PSP, a Secure Trusted Execution Mode on each core). The PSP
hardware is inherently trusted by the platform. Among other things, this
security processor typically responds to requests to perform certain
operations (e.g., store a record of the measured boot sequence for later
enquiry, provide secure storage for asymmetric or symmetric keying
material, provide key access protection, provide encryption and decryption
of small amounts of secure material, provide an internal True Random
Number Generator (TRNG), etc.). However, these security processors are
generally not virtualized so that access from an OS running in a virtual
machine is not possible – the security processor typically assumes a
one-to-one relationship with the OS (e.g., the TPM provides only a single
owner, a single Storage Root Key and password, a single Endorsement Key,
and a single set of PCRs), so that the security processor functions are only
available to the hypervisor in a virtual environment. That means that even
though an OS executing in a virtual machine is capable of utilizing the
secure storage and cryptographic functions of the security processor, as it
would running in a bare-metal environment, it is not able to do so.
The solution to this lies in the implementation of a virtual security processor

(e.g., vTPM, vPSP) that allow a large number (preferably limited only by
available resources rather than design) of virtual machines to maintain a
one-to-one relationship with the virtual Platform Security Processor (vPSP)
that is allocated to it. A virtual implementation of a vPSP allows both the
VMs and the PSP to be unaware of the virtualization and they can both look
and act as they do in a non-virtualized environment. It is necessary to
implement to full interface capabilities of the PSP in a vPSP as well as
protected management functions to create and destroy vPSPs. The software
to implement the vPSP must be integrated into both the hypervisor and the
guest. In the guest a proxy driver is needed to field the API calls and pass
them to the hypervisor component where they can be authenticated – i.e.,
can this VM access that object? The proxy in the guest must also return the
result to the requester. The behavior of the vPSP in the guest must be the
same as the behavior of the physical PSP when it is presented with a
request. The requests can be redirected to the physical PSP or emulated, as
required. The vPSP can protect platform configuration, measurement data
and provide attestation with regard to the state of the platform
configuration, data protection for the OS and its applications, and can help
facilitate remote attestation. The implementation should ensure that
implementations that work with the current PSP continue to work with the
vPSP.
125
PSP operations should never occur in the performance path of any software
implementation. Therefore, the additional latency imposed by a virtual
implementation should be minimized.
10.1.1.2 True Random Number Generator (TRNG)
A TRNG extracts randomness (entropy) from a physical source of some type

and then uses it to generate random numbers. The physical source is also
referred to as an entropy source.
Almost all cryptographic protocols require the generation and use of secret
values that must be unknown to attackers. For example, random number
generators are required to generate public/private key pairs for asymmetric
(public key) algorithms including RSA, DSA, and Diffie-Hellman. Keys for
symmetric and hybrid cryptosystems are also generated randomly. RNGs are
used to create challenges, NONCE (salts) values.
Because security protocols rely on the unpredictability of the keys they use,
random number generators for cryptographic applications must meet
stringent requirements. The most important property is that attackers,
including those who know the RNG design, must not be able to make any
useful predictions about the RNG outputs.
The major use for hardware random number generators is data encryption,
for example to create random cryptographic keys to encrypt data. They are
a more secure alternative to pseudorandom number generators (PRNGs) -
software programs commonly used in computers to generate "random"
numbers. PRNGs use a deterministic algorithm to produce numerical
sequences. Although these pseudorandom sequences pass statistical pattern
tests for randomness, by knowing the algorithm and the conditions used to
initialize it, called the "seed", the output can be predicted. Because the
sequence of numbers produced by a PRNG is predictable, data encrypted
with pseudorandom numbers is potentially vulnerable to cryptanalysis.
Hardware true random number generators (TRNGs) produce sequences of
numbers that are not predictable, and therefore provide the greatest
security when used to encrypt data.
Fog systems should implement a TRNG as opposed to a PRNG solution. The

functionality may be implemented as an ISA extension or via a separate
accelerator device for example. If it is a device, it should be hardware
virtualized, to allow for secure access from multiple VMs and/or containers in
order to preserve secure access.
126
10.1.1.3 Secure Key Generation, Encryption and Storage
The PSP may act as a secure vault for certificates, keys and passwords,
negating the need for costly tokens.
10.1.2 Node Security Aspect
Figure 45 OpenFog Node Security Architecture
The previous figure is divided into four horizontal “zones”: First, at the
bottom, is the hardware component layer (including external devices). A
number of optional (depending on use case requirements) hardware
accelerators may be present here. Shown is an encryption device on the SoC
(it may also be an external device or present as special instructions in the
processors ISA). Other generic accelerators are also shown here. The system
127
mmu and iommu (which may be a single implementation or split) are also
located at this level along with the physical cores. The Hardware Root-of-
Trust (HW-RoT) is also a part of the hardware infrastructure and may be
embedded on-chip or in an external device which provides this function.
At the next level up is the system Firmware, Option ROMs, and Platform
NVRAM. The exact nature and existence of these components is platform
dependent. In order to support the HW-RoT and the extension of the Chain
of Trust, there must be an immutable firmware implementation resident on
trusted system ROM that is the first code to execute on the platform after
power on.
Above that is the Hypervisor layer. It instantiates and manages the virtual
device instances, e.g., the vSoC devices shown, and assigns them to the
virtual machines as directed by the OAM (Operations, Administration, and
Management) system. It also instantiates other virtual devices representing
physical external devices (such as the vNICs shown). These virtual devices
may be entirely supported by hardware that bypasses the hypervisor for
data (such as a sr-iov compliant device) or as software emulated virtual
instances (such as a hard disk that is shared). The virtual cores, which may
or may not be hardware threads if SMT (Simultaneous Multi-Threading) is
supported by the physical core, allow for the presentation of additional
virtual cores.
The final layer is the layer where VMs are instantiated. The physical
resources are mapped here as virtual resources by the hypervisor. The OS in
the VM manages the application address spaces which may be instantiated
as separate application address spaces or as [Linux] containers.
There are a number of functions which connect the layers and provide
system services that help create a secure Chain of Trust comprised of
trusted components. These are represented by the vertical arrows between
the layers.
One example is the “Security Engine” that instantiates a Trusted Execution

Environment and provides services to the hypervisor. The hypervisor, in
turn, virtualizes that engine, the vSE – virtual Security Engine shown in the
hypervisor layer with an agent resident in each trusted VM. The other is the
Trusted Boot firmware and software that verify/measure each subsequent
load of firmware or software to establish a Chain of Trust that includes the
VM.
128
The Trusted Boot/Trusted Loader mechanism is meant to ensure that each
successive code load, be it firmware or software, is trusted allowing for the
extension of the Chain of Trust.
Optionally, untrusted software may be instantiated in a VM to create an

untrusted environment. This configuration may be useful for testing
untrusted material in an otherwise trusted environment and is secured using
the isolation and other security mechanisms described previously.
The RTIC mechanism is described further in 10.1.2.1. It exists in the context

of the hypervisor and monitors the state of the areas of memory that should
not be modified during execution.
Around the perimeter of the abstract fog node system diagram pictured in is
a red line used to describe Physical Security and anti-tamper boundaries
implemented by the physical security and anti-tamper mechanisms. We will
start with that discussion and then proceed to the Root-of-Trust discussion
and work outward from there.
10.1.2.1 Run-time Integrity Checking (RTIC) and Introspection
Secure or measured boot do not ensure that the software that has been
securely instantiated is either free of bugs, infection or remains
uncompromised during execution. The intent of Runtime Integrity Checking
(sometimes called [Hypervisor] Introspection) is to monitor and detect
changes to code and static data in the image during execution. This is done
by “understanding” the image construction, i.e., where code and static data
pages are, by running a set of RTIC specific tools over them before
execution. The hypervisor hosts the RTIC mechanism. The underlying
assumption is that the hypervisor itself is trusted. RTIC is only used to check
VMs. The mechanisms used are mostly passive as the page tables are
modified to detect writes to pages that should not be written to. The action
on detecting an unauthorized modification is driven by policy. Typically, the
VM is terminated.
There are no existing product implementations of this approach but at least

one implementation is under development for both the KVM and Xen
hypervisors.
The other approach that can address this problem in part is memory
encryption as discussed previously. This protects the code and data in an
encrypted “container” (not to be confused with Linux containers) from
outside attacks. It still may be possible for bugs to be exploited or for
previously infected images to be compromised. These “containers” are also
129
vulnerable to compromised services when they have to go outside the
container for services or data. While the code and data (static and dynamic)
can be protected in this way, none of the code or data or data outside of the
container(s) are protected.
When a more mature solution to this problem is available, fog nodes should
implement an RTIC approach to protect the node from being compromised.
This does not necessarily prove more useful for nodes in public places than
those in protected areas as the attacks do not necessarily require physical
access.
10.1.2.2 Debug, Performance Monitoring and Profiling Control
All forms of debug (both hardware and software), performance monitoring,

and profiling control should be turned off after system deployment. These
mechanisms provide a way for third parties with either physical access or
remote access (depending on the mechanism) to provide a technique to
either defeat security mechanisms in place or to gain insight into the
behavior of the system that allows for future side-channel attacks.
If debugging or other monitoring or profiling information is required in the

field, then there must be mechanisms in place to ensure secure provisioning
of the authorization for the specific access by legitimate personnel.
10.1.3 Network Security Aspect
As a pervasive computing infrastructure deployed between the OT frontend

devices and the cloud computing data centers, a secure OpenFog platform is
not merely capable of offering highly-available real-time trusted computing
services but also well-positioned to implement dynamic multi-tier defense-
in-depth strategies to protect the cyber-physical systems that are critical to
our daily living. In order to fulfill these dual missions, the OpenFog platform
must augment the enforcement of node security with the provisioning of
Network Security and the support of continuous Security Monitoring and
Management.
130
Figure 46 OpenFog Security Functional Layers and Operational Planes8
The figure above illustrates such an architecture for providing end-to-end

security with its two operational planes: Security Provisioning and Security
Monitoring and Management, and its three functional layers:
Communications Security, Services Security and Applications Security. This
architecture complies with ITU-X.805 Recommendation [X.805] and also
conforms to the Software Defined Networking (SDN) Architecture [ONF/SDN]
recommended by Open Networking Foundation (ONF). Following subsections
discuss the functional layers in greater detail.
10.1.3.1 Communications Security Layer
This layer implements the following communication security services recom-

mended in [X.800] in all the physical/virtual communication channels among
all the entities in the Device-Fog-Cloud Computing Hierarchy.
 Confidentiality
o Connection and Connectionless Data Confidentiality
o Traffic Flow Confidentiality
 Integrity
o Connection Integrity with Recovery
o Connectionless Integrity with Detection
o Anti-replay Protection
 Authentication
o Data Origin Authentication for Connectionless Communications
o Peer Entity Authentication for Connection-based Communications
o Authenticated Channel Access Control
 Nonrepudiation (optional)
o Nonrepudiation of Origins
o Nonrepudiation of Destination
8 The three security layers conform to the reference architecture of both Open Network
Foundation (ONF) and ITU-X.805 recommendation.
131
The communications occurring in the Device-Fog-Cloud Computing
continuum can be categorized into three kinds of Secure Communication
Pathways:
 Node-to-Cloud Secure Communication Pathways
 Node-to-Node Secure Communication Pathways
 Node-to-Device Secure Communication Pathways
Since the fog nodes often function as the proxies of cloud servers towards
their associated frontend devices while aggregating and representing these
frontend devices to the cloud servers, these pathways shall cooperate to
preserve the interoperability among the frontend devices and the cloud
servers. The following paragraphs highlight the expected functions and the
recommended practice of each kind of pathway.
Figure 47 OpenFog Secure Communication Pathways
10.1.3.2 Node-to-Cloud Secure Communication Pathways
In order to secure these Communication Pathways, the fog nodes are

expected to implement all the X.800 communication security services
(including non-repudiation) for themselves and on behalf of the frontend
devices they represent. Strong authentication and non-repudiation services
shall be implemented using security credentials derived from the hardware
root-of-trust installed in the fog node. Channel Access Control shall be
enforced according to the Communication Security Policies established
between the cloud service providers and the fog node managers as a part of
their service level agreements. All cryptographic operations shall be
132
performed by the crypto accelerators embedded in the fog nodes while the
cryptographic keys shall be managed as a part of the security monitoring
and management operation.
These pathways are also expected to preserve the Internet communication

protocols and APIs employed by cloud servers to communicate with the
frontend devices including IoT devices, personal mobile devices, point-of-
sales (POS) terminals, stand-alone computers and servers. Almost all these
communications are currently conducted as web service transactions via the
following two protocol suites.
Applications Transaction Protocols Security Protocols
Enterprise Apps SOAP over HTTP WSS
Mobile/Personal Apps RESTful HTTP/COAP TLS/DTLS
Figure 48 Protocol Suites for Secure Node-to-Node Communications
10.1.3.3 Node-to-Node Secure Communication Pathways
A distributed fog computing platform may consist of a hierarchy of fog nodes

spanning across multiple Internet subnets or administrative domains, and
yet these fog nodes are expected to coordinate with one another to
accomplish specific objectives. Inter-node information interchanges based on
the transaction based client-server computing model and the event based
publish-subscribe messaging patterns shall both be implemented in order to
enable direct and timely interactions. The following protocol suites are
commonly used to implement these paradigms.
Paradigms Transaction Protocols Security Protocols
Client-Server SOAP, WSS,
RESTful HTTP/COAP TLS/DTLS
Publish-Subscribe MQTT, AMQP, RTPS TLS/DTLS
Figure 49 Protocol Suites for Secure Node-to-Node Communications
Like the node-to-cloud pathways, the node-to-node pathways expect the fog
nodes as the communication endpoints to implement all the X.800 commu-
nication security services including non-repudiation. Strong authentication
and non-repudiation services shall be implemented using security credentials
derived from the hardware root-of-trust installed in the fog nodes. Channel
Access Control shall be enforced according to the communication security
policies established among the fog node managers as part of their service
level agreements. All cryptographic operations shall be performed by the
crypto accelerators embedded in the fog nodes while the cryptographic keys
shall be managed by the security monitoring and management operation.
133
10.1.3.4 Node-to-Device Secure Communication Pathways
Often function as the proxies of the cloud servers, the fog nodes communi-
cating are expected to preserve the communication protocols and APIs used
by the frontend devices. Unfortunately, the choices of device communication
protocols are diversified among different applications and communication
media. Efforts on protocol convergence among wireless, powerline
communication and industrial automation have been made through
adaptation of the Internet (TCP/UDP/IP) protocol suite.
Most of the X.800 communication security services (perhaps, excluding non-

repudiation) can be implemented over the wired/wireless Ethernets and on
the Internet network and transport layers by well-known security protocols.
Among the frontend devices adapted to Internet protocols, strong authen-
tication can be implemented using security credentials issued to the frontend
devices. Channel Access Control can be enforced according to the com-
munication security policies specified by the fog service providers. All
cryptographic operations can be performed by the crypto-enabled embedded
processors in the frontend devices while the cryptographic keys can be
managed as a part of the security monitoring and management operation.
However, among many frontend devices that are not Internet savvy and
often resource-constrained, only limited cryptographic capability such as
symmetric ciphers using manually installed keys is available. These devices
must be installed in physically protected environments and connected via
hardware connections to one or more fog nodes that can provide most of the
X.800 communication security services.
As we further investigate fog computing we will continue to expand the

coverage of node-to-device communications.
Layers Protocols
PHY & MAC Layer  WLAN: 802.11
 WPAN: 802.15
 PLC: PRIME
 Automation: CIP
Wireless Protocol  WiFi
Stacks  Bluetooth
 ZigBee
Adaptation Layer  WLAN/WPAN: 6LowPAN
 PLC: PRIME IPv6 SSCS
 Automation: EtherNet/IP
134
Transport/Network  UDP over IPv6
Layers  TCP over IPv6
 uIPv6 Stack
Application Layer  CoAP
(Publish-Subscribe  MQTT
Messaging)  AMQP
 RTPS
Routing  RPL
 PCEP
 LISP (Cisco)
Security  802.1AR – Secure Device Identity
 802.1AE - Media Access Control (MAC) Security
 802.1X – Port-Based (Authenticated) Media
Access Control
 IPsec AH & ESP, Tunnel/Transport Modes
 (D)TLS – (Datagram) Transport Layer Security
Figure 50 Protocol Suites for Secure Node-to-Device Communications
10.1.3.5 Services Security Layer
This layer offers information security services that are provided traditionally
by network security appliances such as the following:
 Deep Packet Inspection (DPI)
 Application Layer Proxy
 Lawful Message Intercept
 Intrusion Detection and Protection Systems (IPS/IDS)
 System/Network Event and State Monitoring
 Content Filtering and Parental Control
It may also offer networking services often bundled with security services
such as:
 vRouters
 WAN Accelerators
 Network Address Translators (NAT)
 Content Delivery Servers
With the increasing use of Software Defined Networking (SDN) implementa-

tions to replace dedicated devices, these “appliances” are increasingly being
135
implemented as software solutions in virtual machines and Linux containers.
Along with other appliances listed above, this category of security appliances
is generally referred to as Network Function Virtualization (NFV) or
individually as Virtual Network Functions (VNFs). These VNFs, along with
other individually packaged services will likely be chained together on a
Service Function Chain (SFC) and will use Network Service Headers to route
the packets in the selected Service Function Path (SFP).
In many cases, it is believed that these Service Functions will be

implemented in OpenFog systems. The NFV and SFC environment present
their own set of security issues and include many approaches already
discussed but also introduce some new challenges discussed below.
Trusted VNF-to-VNF communication that provides and preserves data

integrity and confidentiality requires a number of features from the platform
hardware, firmware, and software. In addition to a chain of trust developed
from a hardware root of trust, the following features will be required:
 Secure key provisioning for VNF + CA basis for establishing identity
o Authentication of the Virtual Network Function (VNFC)
o Asymmetric crypto
 Bulk data encryption
o Symmetric crypto
 Secure persistent key store
o For private keys
 Trusted VNF-to-OAM/MANO communication (Integrity, Confidentiality)
o Secure software update
o (Same provisioning as above)
 Attestation
o Both parties are in a secure state
Additional security considerations are introduced in the areas of:

 Service Overlay: Transport forwarding for SFFs
o Use packet encryption between SFs/VNFs
o SFF must Authenticate SF/VNF endpoints
 Boundaries of SFC-enabled Domain
o Authenticate Trusted Parties at Boundary: prevent Spoofing,
DDoS, etc.
 Classification
136
o Authentication and Authorization of Classification Policy from
OAM
 SFC Encapsulation
o Metadata needs to be authenticated as to origin
o Selective sharing of sensitive metadata: encrypted or
transformed
In addition, the Network Service Header (NSH) provides functionality that

creates dynamic relationships that may not be authenticated ahead of time
and may fall to the Service Function Forwarder (SFF) to implement.
 Any Service Function (SF) or SFF can update the Service Function Path
(SFP) on the fly.
 The SFP can list an SF more than once in the SFP.
In addition:
 The NSH can contain arbitrary metadata fields (fixed or variable
length), added by the original classifier or by the SFs or SFFs as the
SFP is traversed. These are used to communicate context information
that might be useful to other SFs in the chain.
This introduces another data Confidentiality and Privacy condition. Since the
metadata can contain any data that one of the components in the SFP
deems needed, it is also not clear how to selectively hide (or encrypt) some
fields from one SF (when the packet may cross a service provider, customer,
or department boundary where certain information is considered proprietary,
secret, or sensitive) to others in the chain. The architecture has not yet
addressed these issues. It is a complex problem involving dynamic, unknown
parties.
10.1.4 Data Security Aspect
There are three general categories in which data resides in a system:

 In memory during processing
 On some kind of non-volatile memory
 In messages sent and received on network interfaces
10.1.4.1 Data in Use
Data is resident in the memory system hierarchy (e.g., SRAM, DRAM,

caches, swap space, etc.) during processing. Some of this data, such as
137
keying material, personal data, company proprietary data and, in some
cases, even proprietary algorithms, are considered secret and need to be
protected from being read or altered by unauthorized parties.
As already discussed, memory management units (e.g., the mmu, iommu,

smmu) can be used to protect memory from unauthorized access from other
address spaces (such as VMs) and from devices (either physical or
logical/virtual). The read/write/no-execute page attribute bits also provide
restricted access within an address space (this assumes that the OS or
hypervisor managing the hardware resources is also trusted). The hypervisor
can add some additional protection by abstracting and virtualizing hardware
that could directly affect the execution context of another virtual machine.
Memory that is resident on swap space should also be protected. The

objective is to prevent unauthorized parties from reading the data in the
pages on disk by, for example, removing it and reading it on another
system. This can be done using encryption. Whole disk encryption is one
approach to providing this functionality at a relatively low overhead.
Access to memory using external hardware debuggers, such as JTAG, and

software debuggers should not be enablable in production systems in the
field. JTAG should always be turned off when leaving a lab or controlled
environment. When debugging is required, and if permitted, in the field,
controls must be in place to ensure only an authorized user can use the
debug interface and that it is disabled for all other access.
Encrypted Memory
Memory encryption is used to provide Confidentiality of code and data during

execution. It is used to protect secrets in memory even when parts of the
rest of the system have been compromised. Memory encryption is used
based on the fact that only the CPU package is considered trusted – the
memory is not. As a side effect, it also prevents an attacker from injecting
code into a running image as decryption would result in corrupt code and
program failure.
Memory encryption schemes typically use symmetric key cryptography

because of its speed. This functionality requires both additional hardware
support, including an encryption device resident in the memory management
subsystem, some operating system support to manage the encryption
hardware, and a method for managing the keys associated with the
encrypted memory.
138
Memory encryption is not without a cost. The dynamic decryption/encryption
of memory as it is fetched into cache and written back to memory from
cache affects memory response times.
A complete discussion of memory encryption technology is beyond the scope

of this paper. However, the technology has clear security advantages for at
least some classes of data and some applications in a fog computing
environment. It may be implemented where it is justified by the threat
analysis.
10.1.4.2 Data at Rest
Data at Rest refers to data resident on some non-volatile storage, such as

hard disk or, SSD, USB thumb drives, CDs, DVDs, etc. Encryption is the
front-line defense for data at rest. Among other classes of data, it protects
personally identifiable information (Privacy) and other sensitive data
(Confidentiality). It limits access to those with the correct keys, preventing
anyone who doesn’t have the keys from accessing the data. It provides
protection against unauthorized access to data should the storage media
become physically compromised in some way.
It also meets many compliance requirements, removes any concern

regarding retirement of the storage media and obviates the threat of
physical compromise of the data against unauthorized access - even if
someone with physical access walks away with the drive from the fog node,
they will not have access.
Encryption by itself is not sufficient - keys, policies and certificates must be

actively managed in a secure store, making sure that they are not
compromised and do not fall into the wrong hands.
A process for monitoring who, what, where, when and how data is accessed
from within databases, applications, and the OS/file system should be put in
place. Both access to sensitive information and unauthorized access
attempts should be monitored. All security events should be logged for
subsequent forensic analysis and use by the Operations, Administration, and
Maintenance (OAM) system – e.g., it may use the access data to determine
if an attack is underway. The policies are specified by the OAM system.
Secure data at rest mechanisms must be built on a secure chain of trust,

from power-on through the boot phases, through the instantiation of the
hypervisor (if used), and through instantiation of the operating system and
application in the VM (if a virtual environment is used).
139
There are generally three methods of securing and encrypting data at rest:
Full Disk Encryption
Full-disk encryption is typically implemented using a hardware-based

encryption mechanism in the disk firmware although software disk
encryption implementations also exist. It works by automatically encrypting
all data written to the disk and automatically decrypting all data read from
the disk. Both operations depend on having the correct authentication key.
Without the proper authentication key, even if the hard drive is removed it
cannot be read by another machine running the same or different software.
The advantage of full-disk encryption is that it requires no special attention
on the part of software or the OAM system. If software encryption is used,
because everything on the hard drive is encrypted, including the operating
system, the encrypt/decrypt process can increase data access times. Full-
disk encryption will be most useful for fog devices located in publically
accessible locations (e.g., malls, lamp posts, street corners, roadside, in
vehicles, etc.). Because one key is used to encrypt the entire hard drive, the
OAM system should provide an encryption key backup mechanism in case
the system becomes non-functional for some reason and data retrieval is
required. Carefully managed secure backups may also be used.
File System (and Database) Encryption
File system-level encryption provides a means to use a separate key-based

access and authentication mechanism to protect specific files on a file or
directory/folder basis. It is used when individual files stored on disk (or other
media) need to be protected even from other applications (or users) that
have access to a fully encrypted disk. In use files are encrypted using
symmetric File Encryption Key (FEK). The FEK in turn is encrypted using
owner’s public key. The encrypted FEK is stored with the encrypted file. To
decrypt the file, the file system first decrypts the embedded FEK using the
private key that matches the owner’s public key. Then the file is decrypted
using the FEK. Whole databases, or individual records or fields in records
may also be encrypted. File system encryption may be used by applications,
running in the same VM, that consider their data proprietary or for files that
contain otherwise sensitive or private data.
File System Access Control Mechanisms
File system access control mechanisms may be used to restrict access to

specific files or groups of files by userid or groupid. All modern file systems
implement file permissions in some form. For discussion purposes, the Linux
140
file system is used here. Most of this type of access control will be similar in
other systems.
The Basic File Permissions are applied to Permission Groups using Permission
Types. This is not intended to be a complete or extensive discussion of file
system permissions, but is intended for discussion purposes in the
document.
Permission Groups - Each file and directory has three user based permission
groups:
 owner - The Owner permissions apply only the owner of the file or
directory. They do not impact the actions of other users.
 group - The Group permissions apply only to the group that has been
assigned to the file or directory. They do not affect the actions of other
users.
 all users - The All Users permissions apply to all other users on the
system. This is the permission group that is usually most important.
Permission Types - Each file or directory has three basic permission types:
 read - The Read permission refers to a user’s capability to read the
contents of the file.
 write - The Write permissions refer to a user’s capability to write or
modify a file or directory.
 execute - The Execute permission affects a user’s capability to execute
a file or view the contents of a directory.
These mechanisms are important controls within the context of the OS that
defines userids and groupids. Usually an administrator. Operating from the
OAM, will specify access permissions when a userid and/or groupid is set up
for a specific OS file system environment. This may or may not be used in a
given fog system. It may be important if different applications are running in
the same OS (VM) context, each with the need to different access to the
data in a shared file system (e.g., read-only for some, and read-write for the
data producing application.
10.1.4.3 Data in Motion
Data in motion, sometimes known as data in transit, is used here to describe

packets sent and received on a network interface (including virtual network
interfaces) from or to a fog node – i.e., information that is moving through a
network. Encryption should be implemented for all sensitive or private data
141
in motion: using VPNs, SSL and other technologies which can protect data
from being compromised or seen in plaintext form while in transit.
There are two ways to use encryption when trying to protect data in motion:
using an encrypted connection or using an encrypted file.
An encrypted connection is one in which anything that is sent over a network

connection is automatically encrypted, regardless of the encryption status of
the information to be sent. For example, if sending an already encrypted file,
it will get encrypted again (with a different key) while being sent.
Another method of ensuing data in motion is secure during transit is to use

an already encrypted file. Since an encrypted file exists in encrypted form, it
will always be encrypted, and therefore, protected.
142
11 Glossary
Term Definition Source

Access Control Means to ensure that access to ISO/IEC
assets is authorized and restricted 27000:2014
based on business and security
requirements.
Note: Access control requires both

authentication and authorization.
Actuators “An actuator is a mechanical device [Sclater2007]
for moving or controlling a
mechanism or system. It takes
energy, usually transported by air,
electric current, or liquid, and
converts that into some kind of
motion.”
Address An address is used for locating and IOT-A
accessing – “talking to” – a Device,
a Resource, or a Service. In some
cases, the ID and the Address can
be the same, but conceptually they
are different.
Analytics Synthesis of knowledge from NIST Interagency
information. Publication 8401-
1
Appliance A computer appliance is generally a Wikipedia
separate and discrete hardware
device with integrated software,
specifically designed to provide a
specific computing resource.
Application “Software that provides an [ETSI- ETR173]

Software application service to the user. It is
specific to an application in the
multimedia and/or hypermedia
domain and is composed of
programs and data”.
143
Architecture “The fundamental organization of a [IEEE-1471-
system embodied in its components, 2000]
their relationships to each other,
and to the environment, and the
principles guiding its design and
evolution”.
Architecture Work product used to express [ISO/IEC
Description architecture. 42010:2011]
Architecture Conventions, principles and ISO/IEC
Framework practices for the description of 42010:2011
architectures established within a
specific domain of application and/or
community of stakeholders
Architecture ”A high-level, aspirational view of [TOGAF9]
Vision the target architecture.”
Aspiration “Stakeholder Aspirations are [E-FRAME]
statements that express the
expectations and desires of the
various stakeholders for the services
that the final [system]
implementation will provide.”
Authentication Authentication is the process of Nexus IoT
verifying a user’s true identity. This Glossary
may involve the use of one or more
means of proof of identification, also
known as factors, such as PIN codes
and smart cards.
Authorization Granting of rights, which includes [ISO 7498-
the granting of access based on 2:1989]
access rights.
Autonomy The ability of an intelligent system IHMC

to independently compose and
select among different courses of
action to accomplish goals based on
its knowledge and understanding of
the world, itself, and the situation.
Availability Property of being accessible and ISO/IEC
usable upon demand by an 27000:2014
authorized entity.
Business Logic Goal or behavior of a system IOT-A
involving Things serving a particular
144
business purpose. Business Logic
can define the behavior of a single
Thing, a group of Things, or a
complete business process.
Choreography Type of composition whose elements ISO/IEC DIS
interact in a non-directed fashion 18834-1
with each autonomy part knowing
and following an observable
predefined pattern of behavior for
the entire (global) composition.
Collaboration Type of composition whose elements ISO/IEC DIS
interact in a non-directed fashion, 18834-1
each according to their own plans
and purposes without a predefined
pattern of behaviour
Confidentiality Property that information is not ISO/IEC
made available or disclosed to 27000:2014
unauthorized individuals, entity, or
processes
Cloud Or, "The Cloud," is generally used as IoT Guide
shorthand for Cloud Computing. The
name "Cloud" comes from the fluffy
cloud typically used in Visio-style
network diagrams to represent a
connection to the Internet.
Cloud A general term for the delivery of IoT Guide
Computing various hosted services over the
Internet. The "as-a-Service"
moniker is used for cloud services
such as Software-as-a-Service,
Platform-as-a-Service and
Infrastructure-as-a-Service. The
back-end for many IoT devices may
be delivered via the Cloud.
Communication The communication model aims at IOT-A
Model defining the main communication
paradigms for connecting elements.
This model provides a set of
communication rules to build
interoperable stacks, together with
insights about the main interactions
among the elements of the domain
model.
145
Composition Result of assembling a collection of ISO/IEC DIS
elements for a particular purpose 18834-1
Constrained A constrained network is a network IOT-A
Network of devices with restricted capabilities
regarding storage, computing
power, and / or transfer rate.
Controller Anything that has the capability to IOT-A
affect a Physical Entity, like
changing its state or moving it.
Credentials A credential is a record that contains IOT-A
the authentication information
(credentials) required to connect to
a resource. Most credentials contain
a user name and password.
Cryptography Discipline that embodies principles, ISO/IEC 18014-
means, and mechanisms for the 2:2009
transformation of data in order to
hide its information content, prevent
its undetected modification and/or
prevent its unauthorized use
Data-centricity Scalable, real- Object
time, dependable, high- Management
performance and interoperable data Group
exchanges between publishers and
subscribers.
Device Physical entity embedded inside, or IIC

attached to, another physical entity
in its vicinity, with capabilities to
convey digital information from or to
that physical entity
Device Endpoint that enables access to a IIC
Endpoint device and thus to the related
physical entity.
Digital Entity Any computational or data element IOT-A
of an IT-based system.
DIKW Data gathered becomes Information
when stored and retrievable
becomes Knowledge. Knowledge
enables Wisdom for autonomous
IoT.
Discovery Discovery is a service to find IOT-A
146
unknown resources/entities/services
based on a rough specification of the
desired result. It may be utilized by
a human or another service.
Credentials for authorization are
considered when executing the
discovery.
Edge Gateway Endpoint that provides an entry IIC
point into enterprise or service
provider core networks
Element Unit that is indivisible at a given ISO/IEC DIS
level of abstraction and has a clearly 18834-1
defined boundary
Note: An element can be any type of

entity
Endpoint One of two components that either ISO/IEC 24791-
implements and exposes an 1:2010
interface to other components or
uses the interface of another
component.
Enterprise Segment of computing mostly OpenFog
focused at traditional IT and
Industrial IT.
Edge Also referred to as Mesh Computing, IoT Guide
Computing this concept places applications,
data and processing at the logical
extremes of a network rather than
centralizing them. Placing data and
data-intensive applications at the
Edge reduces the volume and
distance that data must be moved.
Fog Computing Fog computing is a system-level OpenFog
horizontal architecture that Consortium
distributes resources and services of
computing, storage, control and
networking anywhere along the
continuum from Cloud to Things,
thereby accelerating the velocity of
decision making. Fog-centric
architecture serves a specific subset
of business problems that cannot be
successfully implemented using only
147
traditional cloud based architectures
or solely intelligent edge devices.
Fog Node The physical and logical network OpenFog
element that implements fog Consortium
computing services. It is somewhat
analogous to a server in cloud
computing.
Gateway A Gateway is a forwarding element, IOT-A
enabling various local networks to
be connected.
Global Storage Storage that contains global IOT-A

information about many entities of
interest. Access to the global
storage is available over the
internet.
Identity Properties of an entity that makes it IOT-A
definable and recognizable.
Industry 4.0 Refers to the fourth industrial Nexus
revolution, following the first
(mechanization of production
through water and steam power),
second (use of electricity for mass
production), and third (use of
electronics and IT for automation).
Experts believe that the fourth
revolutionary leap will entail full
computerization of traditional
industries. A key element of
Industry 4.0 is the Smart Factory
marked by adaptability, resource
efficiency and ergonomics as well as
intelligent processes and
communication. Technological basis
are cyber-physical systems and the
Internet of Things.
Industrial An Internet of things, machines, IIC
Internet computers and people, enabling
intelligent industrial operations
using advanced data analytics for
transformational business
outcomes.
148
Information “An information model is a [AutoI]
Model representation of concepts,
relationships, constraints, rules, and
operations to specify data semantics
for a chosen domain of discourse.
The advantage of using an
information model is that it can
provide sharable, stable, and
organized structure of information
requirements for the domain
context.
The information model is an

abstract representation of entities,
which can be real objects such as
devices in a network, or logical,
such as the entities used in a billing
system. Typically, the information
model provides formalism to the
description of a specific domain
without constraining how that
description is mapped to an actual
implementation. Thus, different
mappings can be derived from the
same information model. Such
mappings are called data models.”
Infrastructure Specific services that are essential IOT-A
Services for any IoT implementation to work
properly. Such services provide
support for essential features of the
IoT.
Internet “The Internet is a global system of [Wikipedia IN]
interconnected computer networks
that use the standard Internet
protocol suite (TCP/IP) to serve
billions of users worldwide. It is a
network of networks that consists of
millions of private, public, academic,
business, and government networks
of local to global scope that are
linked by a broad array of electronic
and optical networking technologies.
The Internet carries a vast array of
149
information resources and services,
most notably the inter-linked
hypertext documents of the World
Wide Web (WWW) and the
infrastructure to support electronic
mail.
Most traditional communications

media, such as telephone and
television services, are reshaped or
redefined using the technologies of
the Internet, giving rise to services
such as Voice over Internet Protocol
(VoIP) and IPTV. Newspaper
publishing has been reshaped into
Web sites, blogging, and web feeds.
The Internet has enabled or
accelerated the creation of new
forms of human interactions
through instant messaging, Internet
forums, and social networking sites.
The Internet has no centralized

governance in either technological
implementation or policies for
access and usage; each constituent
network sets its own standards.
Only the overreaching definitions of
the two principal name spaces in
the Internet, the Internet-protocol
address space and the domain-
name system, are directed by a
maintainer organization, the
Internet Corporation for Assigned
Names and Numbers (ICANN). The
technical underpinning and
standardization of the core protocols
(IPv4 and IPv6) is an activity of the
Internet Engineering Task Force
(IETF), a non-profit organization of
loosely affiliated international
participants that anyone may
associate with by contributing
technical expertise.”
150
Internet of The digital network is soon going to Nexus
Things (IoT) connect physical objects (“things”),
persons, machines, devices and
processes. It is expected that 50
Billion devices will be connected to
the Internet by 2020. Contrary to
the Internet as we know it, where
only persons have digital identities,
the Internet of Things equips
physical objects with digital
identities. The objects are
embedded with software, electronics
and sensors that allow them to
communicate with other objects or
persons in the digital or physical
world. IoT will transform all
industries – it is expected that the
new connectivity will set off
automation in almost all fields of
business. Establishing secure
infrastructures and trustworthy
identities is vital for the successful
deployment of this new kind of
network.
Interoperability The ability to share information and [TOGAF 9]
services. The ability of two or more
systems or components to exchange
and use information. The ability of
systems to provide and receive
services from other systems and to
use the services so interchanged to
enable them to operate effectively
together.
IoT Service Software component enabling IOT-A
interaction with resources through a
well-defined interface. Can be
orchestrated together with non-IoT
services (e.g., enterprise services).
Interaction with the service is done
via the network.
Local Storage Special type of resource that IOT-A
contains information about one or
only a few entities in the vicinity of
151
a device.
LTE Long Term Evolution commonly
used in 4G.
Microservices Microservices can be considered a Wikipedia
specialization or extension
of service-oriented
architectures (SOA) used to
build distributed software systems.
As with SOA, services in a
microservice architecture
are processes that communicate
with each other over a network in
order to fulfill a goal. Also, like SOA,
these services use technology-
agnostic protocols. The
microservices' architectural style is
a first realization of SOA that
followed the introduction
of DevOps and is becoming more
popular for building continuously
deployed systems. SOA is more
focused on reusability and
segregation whereas microservices
focus on replacing a large
application(s), with a system that
can incrementally evolve and is
easier to manage.
Middleware Middleware is Wikipedia

computer software that provides
services to software
applications beyond those available
from the operating system. It can
be described as "software glue".
Middleware makes it easier
for software developers to
implement communication
and input/output, so they can focus
on the specific purpose of their
application.
Mobile Edge A standard mostly concerned with MEC
152
Computing equipping computational resources
(MEC) at or near base stations in mobile /
cellular networks
Modularity A property of network elements OpenFog
where individual capabilities can be Consortium
added or removed without
substantial impact of other
components.
Multi-tenancy Software Multitenancy refers to a Wikipedia
software architecture in which a
single instance of a software
application runs on a server and
serves multiple tenants. A tenant is
a group of users who share a
common access with specific
privileges to the software instance.
With a multitenant architecture, a
software application is designed to
provide every tenant a dedicated
share of the instance including its
data, configuration, user
management, tenant individual
functionality and non-functional
properties.
Network Resource hosted somewhere in the IOT-A
resource network, e.g., in the cloud.
On-device Resource hosted inside a Device IOT-A
Resource and enabling access to the Device
and thus to the related Physical
Entity.
On-Premises On-premises software (sometimes
Software abbreviated as "on-prem") is
installed and runs on computers on
the premises (in the building) of the
person or organization using the
software, rather than at a remote
facility such as a server farm or
cloud.
Operational Operational Technology (OT) is the Wikipedia

Technology use of computers (or other
processing devices) to monitor or
153
alter the physical state of a system,
such as the control system for a
power station or the control
network for a rail system. The term
has become established to
demonstrate the technological and
functional differences between
traditional IT systems and Industrial
Control Systems environment, the
so-called "IT in the non-carpeted
areas".
Orchestration Type of composition where one ISO/IEC DIS

particular element is used by the 18834-1
composition to oversee and direct
the other elements.
Note: the element that directs an

orchestration is not part of the
orchestration.
Private Cloud Private cloud is cloud infrastructure Wikipedia
operated solely for a single
organization, whether managed
internally or by a third-party, and
hosted either internally or
externally.
Reference A Reference Architecture (RA) is an IOT-A
Architecture architectural design pattern that
indicates how an abstract set of
mechanisms and relationships
realizes a predetermined set of
requirements. It captures the
essence of the architecture of a
collection of systems. The main
purpose of a Reference Architecture
is to provide guidance for the
development of architectures. One
or more reference architectures
may be derived from a common
reference model, to address
different purposes/usages to which
the Reference Model may be
targeted.
154
Reference A reference model is an abstract [OASIS-RM]
Model framework for understanding
significant relationships among the
entities of some environment. It
enables the development of specific
reference or concrete architectures
using consistent standards or
specifications supporting that
environment. A reference model
consists of a minimal set of unifying
concepts, axioms and relationships
within a particular problem domain,
and is independent of specific
standards, technologies,
implementations, or other concrete
details. A reference model may be
used as a basis for education and
explaining standards to non-
specialists.
Reliability Ability of a system or component to ISO/IEC
perform its required functions under 27040:2015
stated conditions for a specified
period of time.
Resilience The condition of the system being IIC
able to avoid, absorb and/or
manage dynamic adversarial
conditions while completing
assigned mission(s), and to
reconstitute operational capabilities
after casualties.
Resource Computational element that gives IOT-A
access to information about or
actuation capabilities on a Physical
Entity.
Requirement A quantitative statement of [TOGAF9]
business need that must be met by
a particular architecture or work
package.
Scalability A property of networks where their OpenFog
capabilities can grow or shrink Consortium
without undue expense of loss of
efficiency
Sensor A sensor is a special Device that IOT-A
155
perceives certain characteristics of
the real world and transfers them
into a digital representation.
Security The correct term is 'information [ISO27001]
security' and typically information
security comprises three
component parts:
▪ Confidentiality. Assurance that

information is shared only among
authorized persons or
organizations. Breaches of
confidentiality can occur when data
is not handled in a manner
appropriate to safeguard the
confidentiality of the information
concerned. Such disclosure can
take place by word of mouth, by
printing, copying, e-mailing or
creating documents and other data
etc.;
▪ Integrity. Assurance that the

information is authentic and
complete. Ensuring that information
can be relied upon to be sufficiently
accurate for its purpose. The term
'integrity' is used frequently when
considering information security as
it represents one of the primary
indicators of information security
(or lack of it). The integrity of data
is not only whether the data is
'correct', but whether it can be
trusted and relied upon;
▪ Availability. Assurance that the

systems responsible for delivering,
storing and processing information
are accessible when needed, by
those who need them.
Service Services are the mechanism by [OASIS-RM]
which needs and capabilities are
brought together
156
Smart Gateway A Gateway is a forwarding element, OpenFog
enabling various local networks to Consortium
be connected. A Smart (or
Intelligent) Gateway additionally
provides more resources for local
(edge) computing. These resources
can include middleware,
microservices and applications. As
such, a Smart (or Intelligent)
Gateway begins to resemble a fog
Node, as a network element that
provides some fog computing
services. Smart Gateways and fog
Nodes are thus also Appliances.
Storage Special type of Resource that stores IOT-A
information coming from resources
and provides information about
Entities. They may also include
services to process the information
stored by the resource. As Storages
are Resources, they can be
deployed either on-device or in the
network.
System A collection of components [IEEE-1471-2000]
organized to accomplish a specific
function or set of functions.
Thing Generally speaking, any physical IOT-A
object. In the term ‘Internet of
Things’ however, it denotes the
same concept as a Physical Entity.
Unconstrained An unconstrained network is a IOT-A
Network network of devices with no
restriction on capabilities such as
storage, computing power, and / or
transfer rate.
View The representation of a related set [TOGAF 9]
of concerns. A view is what is seen
from a viewpoint. An architecture
view may be represented by a
model to demonstrate to
stakeholders their areas of interest
in the architecture. A view does not
have to be visual or graphical in
157
nature.
Viewpoint A definition of the perspective from [TOGAF 9]
which a view is taken. It is a
specification of the conventions for
constructing and using a view
(often by means of an appropriate
schema or template). A view is
what you see; a viewpoint is where
you are looking from - the vantage
point or perspective that
determines what you see.
Virtual Entity Computational or data element IOT-A
representing a Physical Entity.
Virtual Entities can be either Active
or Passive Digital Entities.
Wireless Wireless communication is the [Wikipedia WI]
communication transfer of information over a
technologies distance without the use of
enhanced electrical conductors or
"wires". The distances involved may
be short (a few meters as in
television remote control) or long
(thousands or millions of kilometers
for radio communications). When
the context is clear, the term is
often shortened to "wireless".
Wireless communication is
generally considered to be a branch
of telecommunications.
Wire line A term associated with a network or [setzer-
communication terminal that uses metallic wire messtechnik2010]
technologies conductors (and/or optical fibers)
for telecommunications.
Wireless Wireless sensor and actuator [OECD2009]
Sensors and networks (WSANs) are networks of
Actuators nodes that sense and, potentially,
Network control their environment. They
communicate the information
through wireless links enabling
interaction between people or
computers and the surrounding
environment.
158
References
[IOT-A] EU IOT-A Terminology.
Online at: http://www.iot-a.eu/public/terminology/copy_of_term
[AIMglobal] Association for Automatic Identification and Mobility.
Online at: http://www.aimglobal.org/
[AutoI] Information Model, Deliverable D3.1, Autonomic Internet (AutoI)

Project.
Online at: http://ist-autoi.eu/autoi/d/AutoI_Deliverable_D3.1_-

_Information_Model.pdf
[CCSDS 312.0-G-0] Information architecture reference model.
Online at: http://cwe.ccsds.org/sea/docs/SEA-

IA/Draft%20Documents/IA%20Reference%20Model/ccsds_rasim_20060308.
pdf
[COMPDICT-M2M] Computer Dictionary Definition
Online at: http://www.yourdictionary.com/computer/m2-m
[E-FRAME] E-FRAME project, available.
Online at: http://www.frame-online.net/top-menu/the-architecture-

2/faqs/stakeholder-aspiration.html
[EPCglobal] EPC Global glossary (GS1).
Online at:
http://www.epcglobalinc.org/home/GS1_EPCglobal_Glossary_V35_KS_June_
09_2009.pdf
[ETSI-ETR173] ETSI Technical report ETR 173, Terminal Equipment (TE);

Functional model for multimedia applications.
159
Online at:
http://www.etsi.org/deliver/etsi_etr/100_199/173/01_60/etr_173e01p.pdf
[ETSI TR 102 477] ETSI Corporate telecommunication Networks (CN);

Mobility for enterprise communication.
Online at:
http://www.etsi.org/deliver/etsi_tr/102400_102499/102477/01.01.01_60/tr
_102477v010101p.pdf
[IEEE-1471-2000] IEEE 1471-2000, “IEEE Recommended Practice for

Architectural Description of Software-Intensive Systems”
[ITU-IOT] the Internet of Things summary at ITU.
Online at:
http://www.itu.int/osg/spu/publications/internetofthings/InternetofThings_s
ummary.pdf
[ISO/IEC 2382-1] Information technology -- Vocabulary -- Part 1:

Fundamental terms
Online at:
http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csn
umber=7229
[ISO 27001] ISO 27001: An Introduction to Information, Network and

Internet Security
[OGS] Open GeoSpatial portal, the OpenGIS abstract specification Topic 12:
the OpenGIS Service architecture.
Online at: http://portal.opengeospatial.org/files/?artifact_id=1221
[OASIS-RM] Reference Model for Service Oriented Architecture 1.0
Online at: http://docs.oasis-open.org/soa-rm/v1.0/soa-rm.pdf
[OECD2009]: “Smart Sensor Networks: Technologies and Applications for

Green Growth”, December 2009.
Online at: http://www.oecd.org/dataoecd/39/62/44379113.pdf
160
[Sclater2007] Sclater, N., Mechanisms and Mechanical Devices Sourcebook,
4th Edition (2007), 25, McGraw-Hill
[setzer-messtechnik] setzer-messtechnik glossary, July 2010.
Online at: http://www.setzer-messtechnik.at/grundlagen/rf-

glossary.php?lang=en
[TOGAF9] Open Group, TOGAF 9, 2009
[Wikipedia IN] Internet page on Wikipedia, online at:

http://en.wikipedia.org/wiki/Internet
[ROZANSKI2005] Software Architecture with Viewpoints and Perspectives.
Online at: http://www.viewpoints-and-perspectives.info/doc/spa191-

viewpoints-and-perspectives.pdf
[Wikipedia WI] Wireless page on Wikipedia.
Online at: http://en.wikipedia.org/wiki/Wireless
[IoT Guide] Internet of Things Guide.
Online at: http://internetofthingsguide.com/d/cloud.htm
[Nexsus] Nexus IoT Glossary.
Online at: https://www.nexusgroup.com/en/glossary/?letter=C
[UF IoT] Universal Framework IoT Glossary.
Online at: https://universalframeworks.com/industrial-internet-of-things-i-

iot-glossary/
[AutoI] Information Model, Deliverable D3.1, Autonomic Internet (AutoI)

Project.
Online at: http://ist-autoi.eu/autoi/d/AutoI_Deliverable_D3.1_-

_Information_Model.pdf
[TOGAF9] Open Group, TOGAF 9, 2009
[Ref-a] - http://csrc.nist.gov/publications/fips/fips140-2/fips1402.pdf
161
[Ref-b] - http://csrc.nist.gov/publications/fips/fips140-2/fips1402annexa.pdf
[Ref-c] - http://csrc.nist.gov/publications/nistpubs/800-131A/sp800-
131A.pdf
Mobile Edge Computing: http://www.etsi.org/technologies-

clusters/technologies/mobile-edge-computing
Industrial Internet Consortium: http://www.iiconsortium.org/
Open Connectivity Foundation: http://openconnectivity.org
162

OpenFog Reference Architecture 2 09 17 PDF

Uploaded by

Copyright:

Available Formats

OpenFog Reference Architecture 2 09 17 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

OpenFog Reference Architecture 2 09 17 PDF

Uploaded by

Copyright:

Available Formats

OpenFog Reference Architecture

for Fog Computing

Produced by the OpenFog Consortium Architecture Working Group

Published February 2017.

This is an OpenFog Consortium document and is to be used in accordance with the

Fog computing provides the missing link in the cloud-to-thing continuum.

A horizontal, system-level architecture that distributes

Fog computing is an extension of the traditional cloud-based computing

OpenFog Reference Architecture Overview

We are guided by the OpenFog Board of Directors and the OpenFog

The OpenFog Reference Architecture (OpenFog RA) is intended to help

Many different technical workgroups in the Consortium are responsible for

For further information on these groups or how to participate, please

OpenFog Reference Architecture Content

The OpenFog RA is part of a suite of documents under construction by the

The OpenFog RA is divided into the following chapters:

The Internet of Things, Cloud and the OpenFog RA

The Internet of Things (IoT) is driving business transformation by connecting

 Connected devices are creating data at an exponentially growing rate,

Unfettered cloud-only architectural approaches cannot sustain the projected

Figure 1 Unfettered Cloud Computing

The OpenFog Consortium invites open participation from across industry,

Figure 2 OpenFog Consortium and Other Consortia

The OpenFog RA enables fog-cloud and fog-fog interfaces. OpenFog

 Security: Additional security to ensure safe, trusted transactions

This document describes a set of high-level attributes of fog computing that

Platform as a service (PaaS) is a category of cloud computing services that

The OpenFog RA describes a generic fog platform that is designed to be

Transportation Scenario: Smart Cars and Traffic

Figure 3 OpenFog Transportation: Smart Car and Traffic Control System

Things include roadside sensors (infrastructure) and on-vehicle sensors.

Fog computing nodes in the vehicle

In-vehicle fog nodes provide services including infotainment, advanced

The Transportation Fog Network

Traffic control systems

Visual Security and Surveillance Scenario

Surveillance and security cameras are being deployed worldwide. These

Smart cities, smart homes, retail stores, public transportation,

City-scale deployments that include placing cameras on traffic lights and

Additionally, privacy concerns must be addressed when using a camera as a

Smart Cities Scenario

Figure 4 Opportunities for Smart Cities

The OpenFog RA enables greater efficiency and economic realities of smart

 Intelligent city with smart parking, shopping, and infrastructure.

Safety and Security

3.3.1 Smart Buildings

Smart buildings may contain thousands of sensors to measure various

Some of this processing and response is extremely time-sensitive. For

A fog node could be responsible for:

 Performing emergency monitoring and response functions.

Additional Use Cases

Fog computing is relevant to many more use cases in addition to those

The following sections describe each pillar of the architecture.

Figure 5 Pillars of OpenFog

Security in the OpenFog RA is not a one-size-fits-all architecture. Rather, it

Security implementations have many different descriptions and attributes

Conformance to the OpenFog RA requirements will ensure that an OpenFog

In many applications, particularly for brownfield deployments, or for tiny

As more complex topologies are created in FaaS implementations, the