White Paper

AWS vs. Azure vs. GCP

Determining Your Optimal Mix of Clouds

Introduction
Cloud vendor competition is heating up. 2015 has marked the shift of cloud
services from early to mainstream adoption, especially within large
enterprises. While AWS remains the clear leader in the market (as it has been
since its inception in 2006), Azure is the fastest growing cloud provider, with
triple digit growth in both 2014 and 2015. Google Cloud Platform (GCP),
though far behind in market share, is still considered a top visionary by
Gartner based on the comprehensiveness of their offering, go-to-market
strategy, enhanced performance and global infrastructure.

Gartner 2015 Magic Quadrant for IaaS, with top three leaders circled
Source: Gartner

www.cloudyn.com | @cloudyn_buzz | cloudyn.com/blog

 White Paper

To better assist in the decision making process, this paper offers a concise
breakdown of the three market leaders: Amazon Web Services (AWS), Google
Compute Platform (GCP) and Windows Azure (Azure), who, according to
Gartner, together hold the majority of the IaaS market in 2015.
Produced by Cloudyn, this eBook provides a high-level overview of the “Big 3”
cloud giants, pointing out the strengths of each provider in both operational
and financial terms. Towards the end, we will also explain why it’s not a
question of either/or, but rather which/how much by examining the benefits
that can be reaped by employing a multi-cloud strategy in the enterprise.

Cloud vendors: features and performance

Upon first glance, all cloud providers seem have the same offering: compute,
storage, networking and other platforms as a service (databases,
microservices, big data, APIs etc.). While this 40,000-foot view is true, it’s a
misrepresentation of the myriad services and differentiation each cloud
provider has created in these categories. In this section, we will examine the
offering of the three providers under each category while pointing out key
differentiators.

Global Infrastructure Deployment

As we all know, cloud services don’t actually appear out of thin air (or cloud).
They are anchored in hyperscale data centers spanning between 100,000-
1,000,000 square feet each (2-20 football fields), where tens of thousands of
servers reside in densely populated racks, along with storage and networking
appliances, as well as advanced systems for supplying power and HVAC to
these data crunching behemoths.

Amazon has deployed data centers across the globe, as seen in their
deployment map below. Each circle represents an AWS “region” (currently 12
regions with five more planned for 2016), which is then divided into
“Availability Zones” (AZ), each of which is in reality a stand-alone data center
(represented by the number in the circle). AZs are located far enough from
each other so that the failure of an AZ doesn’t affect the others, and close

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 2
 White Paper

enough for zero-latency connectivity between them, giving users the sense of
“one big data center” in each region.

AWS Regions and Availability Zones in each region. Source: AWS

Microsoft has been quickly building more and more data centers all over the
world in an effort to catch up with Amazon’s vast geographical presence.
From six regions in 2011, they currently have 22 regions, each of which
contains one or more data centers, with five additional regions planned to
open in 2016. While Amazon was the first to open a region in China,
Microsoft preceded to open the India region at the end of 2015.

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 3
 White Paper

MS Azure Regions. Source: Microsoft

Google has the smallest footprint of the three providers, with four regions,
comprised of 3-4 “zones” (data centers) each. Other data centers provide
regional support against zonal failures and act as redundancy only. Google
makes up for its geographical shortcomings with its global network
infrastructure, which provides high-speed, low-latency connectivity between
its data centers, both on a regional and interregional level (compared to
public Internet connectivity with Amazon and Microsoft), as well as a large
number of its own PoPs deployed in over 30 countries.

Google Cloud Platform Regions. Source: Google

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 4
 White Paper

Compute
One of the staples of IT is compute resources, or servers. In the cloud,
physical (bare-metal) servers, in a process called “virtualization” are broken
down into virtual machines (VMs), each of which acts as a server on its own.
Each bare-metal server is equipped with software known as “hypervisor”,
serving as the regulator of physical resources (CPU, RAM, network bandwidth
etc.) between the VMs.

Amazon, with its Elastic Compute Cloud (EC2), has the largest flavor (server
configuration) variety of all three providers. Its virtual machines are called
“instances” and are divided into nine “instance families”, each of which serves
a different purpose, with 2-5 instance sizes (typically 5) within each family.
The families include general-purpose computing, CPU-optimized, RAM-
optimized, storage-optimized and GPU-optimized families.

Microsoft on one hand has less variety in VM families compared to AWS, but
on the other hand, much more flexibility with regards to machine size. Its
families include general-purpose, optimized machines (better CPU, more RAM
and more SSD storage), performance-optimized (even more than “optimized”)
and network-optimized (32Gbps Infiniband networking).

MS Azure VM families. Source: Microsoft

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 5
 White Paper

On the surface Google’s offering seems to have the least amount of VM

families under its Google Compute Engine (GCE). The three families are general-
purpose, CPU-optimized and RAM-optimized, with 5-6 sizes within each family.
However, with their “custom machines” offering, they can be viewed as having
the largest variety of VM families. This is achieved since the “custom
machines” offering allows the user to define exactly how much CPU-power
and how much RAM she requires on her machine. While custom machines
provide the highest flexibility in choosing the right machine, one must also
take into account the complexity involved in managing endless machine
types. Local or ephemeral storage is also configurable, whereas with Amazon
and Microsoft it’s defined in the machine spec.

GCP gives users the option to define their own custom machine configuration. Source: Google

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 6
 White Paper

Storage
Storage is another basic building block of IT. We will discuss storage service
offerings from the “big three” cloud providers with regards to the two main
types of storage: Block and Object storage.

Amazon’s block storage service is called “Elastic Block Storage” (EBS) and
supports three types of persistent disks: Magnetic, SSD and SSD with
provisioned IOPS. Maximum volume sizes range from 1TB for magnetic disks,
up to 16TB for SSD disks.

Their world-renowned object storage service is “Simple Storage Service” (S3),

with four different SLAs: standard, standard - infrequent access, reduced
redundancy and Glacier (for archiving). All data is stored in one availability
zone, unless manually replicated across AZs or regions.

Microsoft’s storage services are all referred to as Blobs. Page Blobs and Disks
are Azure’s block storage service. It can be sourced as standard (magnetic) or
as Premium (SSD), with volumes of up to 1TB.

Block Blobs (not to be confused with block storage) is Azure’s object storage
service. Similar to Amazon, it’s offered in four different SLA levels: Locally
redundant storage (LRS) where redundant copies of the data are stored within
the same data center; zone redundant storage (ZRS), where redundant copies
are stored in different data centers within the same region; and
geographically redundant storage (GRS) which performs LRS on two distant
data centers, for the highest level of durability and availability.

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 7
 White Paper

A completely different Blob

With Google, storage is organized a bit differently than the former two. Block
storage does not get a category on its own, but is rather offered as an add-on
to instances within GCE. There are two options for either magnetic or SSD
volumes; however the IOPS count is fixed (compared to provisioned IOPS
with AWS). Ephemeral (local) disks are fully configurable and are part of the
block storage offering.

Object storage is called Google Storage, and divided into three classes:
Standard, Durable Reduced Availability for less critical data (similar to RRS in
S3) and nearline, which is for archives, but contrary to Amazon’s Glacier, data
starts streaming within a few seconds, not hours - more like S3 standard -
infrequent access.

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 8
 White Paper

Cloud vendors: Billing and Pricing

After examining the different services offered by the cloud providers, we’ll
look into how they price their services. There are clear, inherent differences
in the way the “big three” organize resources into accounts, and in the way
they price their main services, specifically compute services.

Account and Billing Structure

Amazon organizes resources under accounts. An account is a single billing
unit under which cloud resources may be provisioned. Organizations with
multiple AWS accounts, however, would like to receive one consolidated bill
and not multiple separate bills. AWS enables this by creating consolidated
billing. One of the accounts is designated as consolidated account and all
other accounts are linked to it, hence linked accounts. The bill is then
consolidated to include billing for all linked accounts and the consolidated
account, which together are called consolidated billing account family.

AWS Account Structure

Microsoft employs a hierarchical approach towards accounts management.

The subscription is the lowest in the hierarchy, and the only one, which
actually provisions and consumes resources. An account manages multiple
subscriptions. This may sound similar to the AWS account structure, however
Azure accounts are management entities and don’t actually consume
resources themselves. For organizations without MS Enterprise Agreements

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 9
 White Paper

(EA), this is where the hierarchy ends. Those who have EAs, can enroll their
EA in Azure, and manage all of the accounts under them, with optional cost
center and department administrative hierarchies.

Azure Account Structure

Google employs a flat hierarchy. The different resources are grouped under
“Projects” (which are similar to AWS accounts or Azure subscriptions). There is
no higher entity than projects, however multiple projects can be grouped
under a “consolidated billing account”, similar to AWS consolidated billing. This
billing account is not a consuming entity though, and cannot provision
services, similar to Azure’s accounts.

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 1
0
 White Paper

Pricing Models and Discounts

Cloud providers offer different pricing models and discounts for their
services. Most of these complex pricing models and discounts revolve
around compute services, while simple bulk discounts are usually used with
all other services. There are two objectives to this. First, the providers find
themselves in a highly competitive market and would like to lock in their
users to long-term commitments. The second involves an interest to
maximize the utilization of their infrastructure, as every hour in which a VM is
unused, represents a pure loss for the providers.

Amazon has the most complex and diversified pricing structure for its EC2
services:
 On-demand: Pay per hour of usage, with 1-hour billing granularity
 Reserved Instances: Commit upfront for 1 or 3 years of usage (payment
for 24/7 usage over the term) in return for a discounted per-hour price
(30-70%). Payment options include:
o All-upfront: Pay for the whole commitment upfront - highest
discount rate
o Partial-upfront: Pay 50-70% of the commitment upfront, and the
rest in monthly installments over the time of the reservation;
slightly lower discount than all-upfront.
o No-upfront: Pay nothing upfront, and monthly installments over
the term of the reservation. Significantly lower discount than all-
upfront.

AWS Reserved Instances provide a positive ROI only beyond a certain rate of usage

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 1
1
 White Paper

 Scheduled Reserved Instances: Commit upfront for 1 year of usage over

recurring daily/weekly/monthly timeframes (as opposed to 24/7).
Discounts amount to 5% during “peak hours” (Mon-Fri) and 10% during
“off-peak hours” (Sat-Sun).
 Spot Instances: AWS puts up its unused capacity for a price bid. The
“market” price is updated every 5 minutes and usually encompasses
discounts of around 90% over on-demand. Amazon however can
terminate these instances at any time with only a 2-minute warning.
 Fixed Duration Spot Instances: Instances which are provisioned for
either one or six hours, at 25%-45% discount over on-demand pricing.

Pricing Model Discount Trade-Off

On-demand - -

Reserved Instance 30% - 70% Upfront commitment

Scheduled RIs 5% - 10% Upfront commitment (smaller than RI)

Can be terminated at any time by AWS
Spot Instance 80% - 90%
with a 2-minute warning
Fixed Duration Spot 25% - 45% Terminated after one or six hours
Summary of AWS pricing models

Microsoft, in sharp contrast to Amazon, has one main pricing model: on-
demand, charged by the minute (compared to by the hour with AWS).
Discounts are only offered for bulk monetary commitments, either through
pre-paid subscriptions, which offer 5% discount on the bill, or through
Microsoft’s Enterprise Agreements (EAs), where much higher custom
discounts can be applied in return for an upfront monetary commitment by
the customer.

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 1
2
 White Paper

Google offers on-demand pricing, charged by the minute, and is the only
provider of the three who doesn’t require upfront commitment for receiving
discounts. Sustained use discount is a pricing model, which retroactively
discounts services, which were extensively used over the period of the billing
month. The method is simple to understand, and no special action need be
taken to enjoy its benefits. However, it does complicate the task of
forecasting a project's cost, as it is unclear until the end of the month.

Sustained Usage Discount

Usage Level
(% out of the month)

0-25% 0%

25%-50% 20%

50%-75% 40%

75%-100% 60%

Google Cloud Platform Sustained Use Discount

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 1
3
 White Paper

Summary and Takeaways

Companies preparing for cloud migration must do their homework to find

the best course for them. Each vendor has its own sweet spot for particular
deployments and it is up to the migrating company to figure out which KPIs
are most important to them and select cloud vendors accordingly. Which one
will fit and integrate with business needs best, as well as how it meets
business’ short term and long-term goals.

As the title of this paper suggests, choice of cloud provider is not an either/or
question. With the growing maturity of public and private cloud platforms
and as IaaS hits massive adoption, enterprises are realizing that relying on a
single cloud provider is not a long lasting option. Issues like higher
availability, vendor lock-in and leveraging competitive pricing push
enterprises to find the optimal mix of clouds for their needs, rather than a
single provider.

COMPANY OVERVIEW
Founded in 2011, Cloudyn is the leader in cloud monitoring and optimization. The
company’s industry award-winning SaaS solution delivers unprecedented insights
into usage, performance, and cost, coupled with custom prescriptive actions for
enhancing performance and reducing cloud spend .With more than 10,000,000
virtual instances monitored Cloudyn helps businesses select the right mix of cloud
vendors, increase operational performance, reduce cloud costs to bring them
under optimum control, and capitalize on customer choice. More than 2,400
customers use Cloudyn’s technology worldwide including F500 industry leaders in
aerospace, infrastructure, consumer online travel services, IT management
consulting, and manufacturing . For more information, interested parties may visit
www.cloudyn.com.

AWS & Azure & GCP – Determining your Optimal Mix of Clouds 1
4