Accepted for presentation in: International Conference on Information and Communication Technology Convergence (ICTC)

Software-Defined Networking driven

Time-Sensitive Networking for
Mixed-Criticality Control Applications
Fabian Kurtz1 , Gösta Stomberg2 , Maı́sa Beraldo Bandeira2 , Jens Püttschneider2 ,
Felix Greiwe, Michael Kaupmann, Christoph Hams, Tim Harnisch, Mey Olivares Tay, Asha Choudhary,
Jorge Ramı́rez Treviño, Abhishek Bhanderi, Apurva Rajashekbhar, Padmashree Vemana, Ahmed Alhanalfi,
Timm Faulwasser2 and Christian Wietfeld1
1 Communication Networks Institute (CNI)
2 Institute for Energy Systems, Energy Efficiency and Energy Economics (IE3)

TU Dortmund University, 44227 Dortmund, Germany

Email: {fabian.kurtz, goesta.stomberg, maisa.bandeira, jens.puettschneider, christian.wietfeld}@tu-dortmund.de
timm.faulwasser@ieee.org

Abstract—Developments such as Industry 4.0, Smart Grids, or Centralized User and Network Configuration (CUC, CNC)
Intelligent Transportation System (ITS) depend on reliable high- serve to manage TSN’s features. Yet, integrating TSN into
performance communications to enable the underlying control existing networks benefits from harnessing established control
algorithms. Nevertheless, in most cases it is not viable to provide
network infrastructures exclusively for mission-critical control planes. Therefore, this work proposes the orchestration of TSN
traffic or just the general use case. Hence, 5G as well as future 6G via Software-Defined Networking (SDN), c.f. Fig. 1. To this
networks entail functionalities such as network slicing to enable end, a SDN controller is enhanced for interfacing with TSN-
the coexistence of mixed-criticality applications on unified net- switches. Thereby, the network topology can be monitored and
works. An approach capable of supporting slicing in the wireline managed centrally. This is evaluated empirically for distributed
domain is Time-Sensitive Networking (TSN). It addresses the
mentioned challenges by enhancing the Ethernet standard with approaches to automatic control of dynamical systems in-
functionalities required for deterministically bounded low laten- cluding communication-demanding Model Predictive Control
cies and preemption of high priority data flows. To facilitate its (MPC). These are deployed on embedded platforms acting as
integration with existing 5G and emerging architectures including agents. A bottleneck in terms of network capacity is introduced
6G and open Radio Access Networks (O-RANs), Software- by forcing traffic flows onto a single link and adding best-effort
Defined Networking (SDN) has emerged for orchestrating the
novel feature set. In this paper, we thus combine TSN and SDN services. This highlights the potential of SDN-driven TSN in
to design an integrated solution. We present experimental results challenging mixed-criticality applications such as those found
on handling communication-demanding control algorithms and within Smart Grids. This work is structured as follows: After
cross traffic simultaneously. Our findings underpin the potential an overview of related works in Sec. II, Sec. III details our
of SDN-driven TSN for mixed-criticality control applications. approach to SDN-driven TSN. Next, Sec. IV discusses the
evaluation scenario, setup and results. Finally, Sec. V draws
I. I NTRODUCTION
conclusions and offers an outlook on future work.
Modern technology increasingly relies on highly auto-
mated systems such as Smart Grids, Intelligent Transportation SDN Controller

System (ITS), or Industry 4.0. Their operation necessitates SCADA

System
TSN Switch

TSN
mission-critical control algorithms which in turn depend on Switch PLC
Agents

reliable communications [1]. Yet, as dedicated networks often 5G/6G

are neither available nor viable, these requirements have to TSN Switch Substation PLC

be met with shared resources. This leads to mixed-criticality Agent

traffic which, e.g., can be realized via 5G [2] network slicing Industry 4.0
[3, 4]. Here, as well as for future open Radio Access Network Control
Agent
5G/6G

(O-RAN) [5] and 6G [6] infrastructures, Time-Sensitive Net- Distribution

Grid

working (TSN) has emerged as an technological enabler. First Roadside

Camera
PLC: Programmable Logic Controller
SCADA: Supervisory Control And
SDN Control Vehicular Data Acquisition
specified via Ethernet [7] amendments, it is standardized in TSN Traffic Smart Grids Systems SDN: Software-Defined Networking
TSN: Time-Sensitive Networking

IEEE 802.1Q [8], offering scheduling features via time slots,

guard bands, frame preemption, etc. It is particularly relevant Figure 1: Software-Defined Networking driven Time-Sensitive
in wireline communications, enabling hard service guarantees. Networking enables mixed-criticality control applications

© 2022 IEEE. Personal use of this material is permitted. Permission from IEEE must be obtained for all other uses,
including reprinting/republishing this material for advertising or promotional purposes, collecting new collected works
for resale or redistribution to servers or lists, or reuse of any copyrighted component of this work in other works.
 II. R ELATED W ORKS III. S OFTWARE -D EFINED N ETWORK DRIVEN
T IME -S ENSITIVE N ETWORKING
While the orchestration of TSN via SDN has been proposed
Next, we recall the considered distributed algorithms for
in related work, several opportunities for further research
automatic control which are drawn upon as time-critical com-
remain. Existing studies can be grouped into three main
munication tasks. Then we detail TSN and our approach.
categories: Those works primarily concerned with specifying
novel architectures for SDN/TSN integration [9, 10], those A. Distributed Control
centered around analytic or simulative means [11–16] and We design controllers for a set S = {1, . . . , S} of dynamical
approaches build on empirical testing environments [17, 18]. systems. The dynamics of subsystem/agent i ∈ S are
Regarding works belonging to the first group, the Yet Another X
ẋi (t) = fij (xj (t), uj (t)), xi (0) = xi,0 , (1)
Next Generation (YANG) data modeling language as described
i∈S
in an amendment to IEEE 802.1Q for interfacing with TSN
devices via network management protocols such as NETCONF yi (t) = hi (xi (t)), (2)
or RESTCONF is used in [9, 10]. While providing important where xi ∈ Rni is the system state, ui ∈ Rmi is the system
contributions to the overall discussion, such works are limited input, yi (t) ∈ Rpi is the system output, and the functions fij :
in terms of results supporting the proposed approaches. Rnj × Rmj → Rni describe dynamic coupling of agents. We
Other studies are built around analytic models, e.g., presen- apply two different control methods to (1). Firstly, we design
ted by Li et al. [11] for stream reservation and bounded networked output-feedback controllers for linear, decoupled,
E2E latencies in avionics. Further papers rely exclusively single-input-single-output systems. The control law is
on simulation, often lacking real-world traffic models. For X
ui (t) = − lij yj (t),
one, the authors of [12] focus on end-to-end latencies for in-
j∈S
vehicular use, while [13] also studies resource usage efficiency
as a function of guard band size. In [14] an open source where lij are the elements of a coupling Laplacian L ∈ RS×S ,
SDN controller, interfacing with TSN via YANG, is coupled cf. [19]. The matrix L is designed to achieve output synchron-
to a simulation implemented in the OMNeT++ framework for ization, i.e., limt→∞ ∥yi (t) − yj (t)∥ = 0 ∀i ∈ S, ∀j ∈ S. This
solving a linear optimization problem. Their focus is placed control method requires communication coupled subsystems to
on assessing various scheduling schemes and their impact exchange messages once per control interval.
on bandwidth utilization. Traffic classification for accurate As an alternative, we employ Distributed Model Predictive
identification and subsequent prioritization of critical data Control (DMPC) (1) to stabilize the systems at xi = 0 ∀i ∈ S.
flows is discussed in [15]. There, OMNeT++ is harnessed To this end, we discretize (1) to obtain discrete-time dy-
to determine end-to-end latencies achievable by the selected namics fij d
(·). Moreover, we consider compact and convex
approach. Building on this work, the same simulation envir- constraints Ui . In DMPC, the agents at sampling instant
onment is also applied in a proximate study by the same tk , k ∈ N measure their current system state x(tk ) and
researchers in [16]. In an effort to ensure a stable performance cooperatively solve a discrete-time Optimal Control Problem
level in highly dynamic network environments, an SDN-based (OCP) with horizon N of form
approach to path reconfiguration is presented. N −1
1 X X k⊤
N⊤
Finally, research based on measurements performed within min xi Qi xki + uk⊤ k
i Ri ui + xi Pi x N
i (3a)
xi ,ui 2
testing setups or even real-world environments, such as dis- i∈S k=0

cussed in this paper, is significantly scarcer. A publication subject to ∀i ∈ S :

X

by Thi et al. [17] facilitates the distribution of a highly xk+1 = d
fij xkj , ukj , ∀k ∈ {1, . . . , N − 1}, (3b)
i
synchronous timing signal throughout industrial networks. j∈S
Observed results mostly remain below 500 ns and as such x0i = xi (tk ), (3c)
are suitable for a wide range of applications in the context
of Industry 4.0 and beyond. Gerhard et al. [18] employ an uki ∈ Ui , ∀k ∈ {1, . . . , N − 1}. (3d)
open source SDN controller to drive TSN for Open Platform The agents then apply the input trajectory’s first part as
Communications Unified Architecture (OPC UA) applications, control input u(tk ) = u0 . OCP (3) is solved by distributed
but unfortunately do not provide any measurement results and decentralized optimization methods to obtain a distrib-
owing to critical implementation issues preventing further uted control scheme. Here, we use the Alternating Direction
evaluation of the presented setup. Method of Multipliers (ADMM) and the distributed Active Set
In contrast, this paper provides an empirical study of Method (ASM) for linear system dynamics and decentralized
SDN/TSN integration on the example of realistic traffic gen- Sequential Quadratic Programming (dSQP) for nonlinear dy-
erated by distributed control algorithms. The impact of cross namics, cf. [20, 21]. For a recent overview of other distributed
traffic on interframe delays of this mission-critical application optimization algorithms applicable to DMPC we refer to [20].
is shown to be mitigated, effectively providing the benefits of Since the applied numerical optimization schemes are iterative
dedicated networks on shared infrastructures, thereby support- in nature, DMPC as a feedback control strategy requires agents
ing network slicing and future 5G/6G communications. to exchange multiple messages per control interval.
 B. Foundations of Time-Sensitive Networking C. Proposed Approach to SDN-driven TSN
TSN is defined by a set of specifications amended to Fig. 3 illustrates our concept for orchestrating TSN via
IEEE 802.1Q [8], in an effort to enable bounded latencies, SDN. Based on the SDN paradigm, the design can be split
i.e., deterministic networking, based on the Ethernet stand- into the three layers of application, control, and data plane.
ard [7] for time-critical applications. Most relevant for this Applications, in this case the highly time-critical distributed
work are the time-aware shaper, time slots and guard bands control algorithms as well as best-effort background traffic,
originally defined in IEEE 802.1Qbv, as well as preemption exist on the top level. They sit alongside user-side interfaces
(IEEE 802.1Qbu). As shown by the top row of Fig. 2, the such as graphical or command line options, and conceptually
time-aware shaper enables scheduling by introducing cyclic can utilize the northbound Application Programming Interface
slots on the Ethernet network, effectively acting as Time- (API) to convey their requirements and commands to the SDN
Division Multiple Access (TDMA). Hence, different traffic controller, e.g. via Representational State Transfer (REST).
classes can be established, allowing the prioritization of critical This approach hence facilitates application aware network
communication versus other flows such as best-effort data. configuration, dynamic communication infrastructure control
Nevertheless, issues may arise as this does not allow the loops and overall highly automated system operation.
interruption of frames currently being transmitted. Building on previous works, we here employ a self-
Thus, should e.g. a best-effort frame arrive within the tail developed SDN controller [22] which in turn is forked from the
end of its assigned slot it can intrude on the slot reserved open source Floodlight project [23]. Aside form a broad range
for critical communications, as illustrated by the blue packet of features and intrinsic network monitoring capabilities, we
above the green slot in Fig. 2 (top row). As such a violation enhanced this entity with support for reconfiguring features
of guarantees is unacceptable, TSN offers guard bands to offered by TSN. This includes setup and parametrization
prevent such situations (c.f. middle row Fig. 2). To ensure of functionalities such as guard intervals, time slots of the
that no packet can be send during guard bands, its length time-aware shaper and preemption. The SDN controller thus
has to equal the maximum frame size allowed. However, this translates commands issued via the northbound API into
approach reduces overall network capacity. Its impact can be concrete configurations. By traversing the southbound API,
minimized via frame preemption as shown in the bottom row traditionally represented by the de-facto standard OpenFlow
of Fig. 2. Here, the ongoing transmission of Ethernet frames protocol, these are transferred to the data plane. At this level,
can be paused at guard bands, resuming once the assigned TSN-enabled Ethernet switches handle the physical forwarding
periodic time slot reappears. Thereby, wasted link capacity is of data packets according to the SDN controller’s directives.
reduced. Additionally, guard bands can be minimized to the This functionality is enabled by way of a corresponding
size of a partial packet, increasing usable data rate even further. module, handling SDN/TSN interaction and implemented in
In terms of configuration topology TSN employs CNC and this work via REST over Secure Shell (SSH) tunnels. Thereby,
CUC. Possible mechanisms such as Management Informa- bounded deterministic latencies and hard prioritization via
tion Base (MIB) via Simple Network Management Protocol time slot scheduling as offered by TSN is orchestrated through
(SNMP) or YANG via NETCONF/RESTCONF are also well our centralized SDN controller. The design also supports 5G
established in Ethernet. Optimal integration into existing com- network slicing, including on the air interface as demonstrated
munication networks can be achieved by building on SDN’s in [24], thus extending these capabilities to TSN.
powerful and proven set of features for network orchestration.
Graphical User Distributed
Best-Effort
& Control
Applications
Time critical time slot Best-effort time slot Command Line Algorithms
Time-critical time slot blocked Interfaces
Application Aware Configuration
IEEE 802.1Qbv

by best-effort frame
Northbound Interface (i.e. Representational State Transfer, REST)
t
Network Guard Time-Aware
Guard bands to avoid frames from Monitoring Interval Setup Shaper
Can not be sent intruding on time-critical time slots
Misc. Features Pre-emption Classification
& IEEE 802.1Qbu IEEE 802.1Qbv

Default SDN TSN Feature Support

Software-Defined Networking (SDN) Controller
t Southbound Interface (here: REST via SSH; default: OpenFlow)
Pre-emption of best-effort frames
Paused frame reduces required guard bands
Agent Agent
IEEE 802.1Qbv

Agent Time-Sensitive Networking (TSN) Agent

t Switches

Cycle n Cycle n+1

Figure 3: Concept for integrating TSN into SDN-driven com-
Figure 2: Overview of Ethernet enhancements introduced by munication infrastructures. The SDN controller monitors the
TSN as used in this work. Time slots and guard bands specified network and translates commands issued via a REST-based
by IEEE 802.1Qbv and 802.1Qbu preemption enable hard northbound API into configuration parameters for the TSN-
service guarantees for mixed-criticality traffic. enabled data plane, thereby enforcing hard service guarantees.
 IV. E VALUATION 100
400 80 max=76.8µs
max=331ns
This section provides an overview of the evaluation scenario
60
and setup, as well as a detailed discussion of achieved results. 200
40

Offset [ns]

Offset [µs]
A. Evaluation Scenario & Experimental Setup 0 20

0
The evaluation setup created for the purposes of this work
-200
is shown in Fig. 4. Two NXP LS1021ATSN serve as data -20

plane, i.e., TSN switches, offering a per link data rate of -400 min=-383ns -40 min=-30.8µs
100 Mbit s−1 . They connect to each other, as well as to -60
0 500 1000 1500 2000 0 300 600 900 1200 1500
embedded computers (i.e., Raspberry Pi 3) acting as dis- Time [s] Time [s]
tributed control agents. However, unlike its peers Agent 4
does not run control algorithms, but introduces generic best- Figure 5: Synchronization of the evaluation setup. Software
effort cross traffic into the network, fully saturating the middle based (control agents, left) vs. hardware based PTP (TSN
link. Thereby, a resource conflict is generated on the central switches, right). Note the different orders of magnitude and
link between both TSN switches. Network orchestration is significantly more stable clocks on the right.
performed by an SDN controller. It is a self-developed solution As control applications, one scenario with the output-
used in previous works [25] and based on the open source feedback controller and three scenarios with DMPC are ana-
Floodlight controller [23]. Among its tasks is the configuration lyzed for networks with three agents (ni = 2, mi = pi = 1).
of the time-aware shaper to ensure hard service guarantees for For the output-feedback controller, we consider a network of
control applications in a mixed-criticality environment. The fully coupled undamped oscillators as individual subsystems
various traffic streams’ dissimilar priority levels are encoded (Scenario 1), design L, and choose a sampling time of 15 ms.
into the field offered by Virtual Local Area Networks (VLANs) In case of DMPC, we design controllers with N = 5 and a
to specify service classes and acted upon by the switches. sampling time of 200 ms for a linear chain of masses system
TSN depends on highly synchronized network entities to en- and apply either ADMM (Scenario 2) or ASM (Scenario 3).
able the time slot based concept required for ensuring bounded We use five ADMM iterations per MPC step to solve OCP (3)
latencies. Therefore, we employ the Precision Time Protocol in Scenario 1 and solve the OCP to optimality with ASM in
(PTP) [26] to establish a firm time reference throughout the Scenario 2. Additionally, we design a controller with N = 4
setup. Hence, timing variance at the different devices is min- and a sampling time of 100 ms for a network of three fully
imized, allowing stable operation which in turn provides the coupled nonlinear Van der Pol oscillators with parameters
high confidence levels required for the subsequent evaluation. from [27] and apply dSQP (Scenario 4). In dSQP, we run
Observed measurement results are given in Fig. 5. As clearly 25 inner iterations per MPC step to solve OCP (3). We use
shown, the TSN switches’ hardware assisted PTP implementa- qpOASES [28] to solve the arising optimization problems
tion yields significantly more homogeneous and precise offsets in ADMM, CasADi [29] to compute sensitivities, and the
in contrast to the software solution employed at the agents. Lightweight Communications and Marshalling (LCM) [30]
While hardware PTP shown on the left mostly exhibits delays library to exchange iteration variables between agents.
in a range of approximately ±100 ns with peaks well into
the 300 ns range, software PTP fluctuates primarily between B. Evaluation Results
±20 µs, with outliers below 100 µs. Taking this into account, Fig. 6 shows the measured interframe latencies for the
results presented in the following are rounded to 100 µs. DMPC scenarios. Each of them is analyzed under three dif-
ferent traffic conditions. Black dots in the subgraphs mark the
Agent 1 Time critical control traffic Best-effort traffic Agent 2
respective mean interframe latencies. The left column displays
scenarios with inactive TSN and no cross traffic. Hence, these
distributions represent the ideal traffic patterns generated by
TSN Switch TSN Switch the employed control algorithms. Any deviation thus implies
unwanted interference with application behavior, i.e., as caused
by a lack of control traffic prioritization. Therefore, the ideal
Resource Conflict
result achievable by the presented SDN-driven approach to
TSN is the avoidance of such variations.
Agent 4 Agent 3 The control applications send messages with an interframe
delay of around 1 ms to solve the OCP. Also, messages with
SDN Controller
larger interframe delays occur when waiting for the following
Figure 4: Setup for evaluating SDN-driven TSN comprised of control interval. This applies in case of ADMM and ASM,
two switches orchestrated by an SDN controller. Four agents whose communication behavior features interframe delays just
execute distributed control algorithms and generate best-effort below 200 ms respectively slightly above 190 ms, as depicted
traffic, thereby creating a resource conflict on the central link. in the corresponding rows of the leftmost column in Fig. 6.
 No Cross Traffic Cross Traffic Cross Traffic 4 4
No TSN No TSN TSN y1 (t) y1 (t)
y2 (t) y2 (t)
(Chain of Masses) (Chain of Masses)

2 y3 (t) 2 y3 (t)
ADMM

Output y

Output y
0 0

−2 −2

−4 −4
ASM

0 20 40 60 0 20 40 60
Time [s] Time [s]

Figure 7: Closed-loop oscillator trajectories with cross traffic

and inactive TSN (left) and active TSN (right). The desired
(Van der Pol)

synchronous oscillation is only possible with active TSN.

dSQP

cross traffic and active SDN-driven TSN. In all cases, inter-

frame delay distributions appear virtually identical to those
Figure 6: Measured interframe latencies for DMPC. Without without competing data flows, as shown on the very left.
TSN (middle row) delay distributions are widened significantly This highlights how TSN ensures the uninterrupted and correct
due to the interference of cross traffic. With TSN active (right), execution of the selected automatic control algorithms despite
performance is identical to a dedicated network (left). resource conflicts. Hence, the viability of our setup and its
underlying technologies for mixed-criticality applications, as
The center column shows results including cross traffic with outlined at the start of this work, is demonstrated and validated
inactive TSN, i.e., as can be observed in traditional Ethernet by way of a realistic application scenario.
based networks. In all three DMPC scenarios, cross traffic Tab. I summarizes the mean interframe delays for the
increases interframe delays thereby interfering with the various analyzed DMPC and traffic scenarios. Results are rounded
algorithms. As a result, the time required to solve the OCP to 100 µs which, as previously discussed, serves to account
also increases, which could deteriorate the control performance for clock jitter highlighted in the Fig. 5 PTP chart. Row one
and thereby impact mission-critical applications. In case of and three present identical values, with a very minor variation
ADMM the distribution widens at slightly above 30 ms and in ADMM caused by the slightest variance in measurement,
around 65 ms, with individual outliers going beyond 200 ms. which is to be expected in an empirical study. In contract,
ASM is not impacted as strongly. Yet, the overall latencies the middle row shows a clear deviation from optimal results
increase visibly with more values scattered up to 190 ms. as caused by interfering best-effort communication, thereby
Lastly, dSQP forms two bulges from 5 ms to 20 ms. Thus, all aligning with the previously discussed delay distributions.
three cases illustrate the distinct effect of cross traffic on timely Hence, this further illustrates the effectiveness of SDN-driven
delivery of traffic generated by distributed control agents as TSN to provide hard service guarantees in challenging mixed-
caused by resource conflicts (c.f. Fig. 4), typically encountered criticality communication environments, e.g. shared networks.
in shared/sliced (public) communication infrastructures. In addition to the DMPC scenarios, we also provide an
Finally, the right column of Fig. 6 presents results with example from output-feedback control to graphically illustrate
the impact of TSN - or lack thereof - on control algorithm
Table I: Mean delays observed in the evaluation of DMPC, performance. Fig. 7 shows the closed-loop trajectories of the
highlighting TSN’s mixed-criticality performance. controlled oscillators for two different network scenarios. On
the right hand side of the plot the desired control performance
ADMM ASM dSQP is shown, i.e., output synchronization of the three agents to
Chain of Masses Chain of Masses Van der Pol
a common oscillation. This control behavior is facilitated by
Mean Mean Mean TSN, which guarantees the timely communication of mes-
Interframe Interframe Interframe
Delay [ms] Delay [ms] Delay [ms] sages relevant for the control algorithm. Performance is thus
effectively identical to what can be measured within a com-
No Cross
Traffic, 15.4 13.4 1.2 munication network dedicated exclusively to mission-critical
No TSN application traffic. Conversely, the left of Fig. 7 shows the
Cross closed-loop result without TSN. Here, cross traffic interferes
Traffic, 23.5 15.6 6.1 with the control application. Thus the desired outcome as
No TSN represented by synchronized oscillation is not achieved. In
Cross summary, this aligns with the performance level expected from
Traffic, 15.5 13.4 1.2 and indicated by the previously discussed network delays.
TSN
This underlines the suitability of our proposed approach to
SDN-driven TSN for supporting and enabling hard service [8] IEEE Standard for Local and Metropolitan Area Network – Bridges
guarantees regardless of interfering traffic. It thus enables and Bridged Networks, IEEE Standard 802.1Q, 2018.
[9] S. B. H. Said, Q. H. Truong and M. Boc, ‘SDN-Based Configuration
wireline network slicing in mixed-criticality use cases within Solution for IEEE 802.1 Time Sensitive Networking (TSN),’ SIGBED
application domains such as Industry 4.0 or Smart Grids. Rev., vol. 16, no. 1, pp. 27–32, 2019.
[10] H. Chahed and A. J. Kassler, ‘Software-Defined Time Sensitive
V. C ONCLUSION AND O UTLOOK Networks Configuration and Management,’ in IEEE Conference on
Network Function Virtualization and Software Defined Networks (NFV-
In this work, we present an approach to SDN-driven TSN SDN), 2021, pp. 124–128.
for mixed-criticality control applications such as Industry 4.0 [11] E. Li, F. He, L. Zhao and X. Zhou, ‘A SDN-based Traffic Bandwidth
or Smart Grids. For this, TSN is integrated with SDN. Thereby, Allocation Method for Time Sensitive Networking in Avionics,’ in
IEEE Digital Avionics Systems Conference (DASC), 2019, pp. 1–7.
novel features introduced to Ethernet by IEEE-defined stand- [12] T. Hackel, P. Meyer, F. Korf and T. C. Schmidt, ‘Software-Defined
ard extensions forming TSN become available for centralized Networks Supporting Time-Sensitive In-Vehicular Communication,’ in
and application aware configuration via SDN. This includes IEEE Vehicular Technology Conference (VTC2019), 2019, pp. 1–5.
[13] H. Lee, J. Lee, C. Park and S. Park, ‘Time-aware Preemption to En-
the time-aware shaper with cyclic slots as well as guard hance the Performance of Audio/Video Bridging (AVB) in IEEE 802.1
bands and frame preemption. A purpose developed testing TSN,’ in IEEE International Conference on Computer Communication
environment is presented and serves to empirically evaluate and the Internet (ICCCI), 2016, pp. 80–84.
[14] M. A. Metaal, R. Guillaume, R. Steinmetz and A. Rizk, ‘Integrated
our approach on the example of challenging traffic generated Industrial Ethernet Networks: Time-sensitive Networking over SDN
by various distributed real-world control algorithms. Here, a Infrastructure for mixed Applications,’ in IFIP Networking Conference
resource conflict between agents performing calculations and (Networking), 2020, pp. 803–808.
[15] N. S. Bülbül, D. Ergenç and M. Fischer, ‘SDN-based Self-
exchanging values across a single central link and generic Configuration for Time-Sensitive IoT Networks,’ in IEEE Conference
best-effort traffic is created. Measurement results focus on on Local Computer Networks (LCN), 2021, pp. 73–80.
interframe latencies to study the impact of cross traffic on [16] ——, ‘Towards SDN-based Dynamic Path Reconfiguration for Time
Sensitive Networking,’ in NOMS IEEE/IFIP Network Operations and
agent communication. Additionally, output trajectories are Management Symposium, 2022, pp. 1–9.
given to visually illustrate the impact of competing packet [17] M.-T. Thi, S. Ben Hadj Said and M. Boc, in IEEE International Con-
flows, or lack thereof, on control. In summary, it is shown ference on Emerging Technologies and Factory Automation (ETFA).
[18] T. Gerhard, T. Kobzan, I. Blöcher and M. Hendel, ‘Software-defined
that TSN allows mixed-criticality environments to achieve a Flow Reservation: Configuring IEEE 802.1Q Time-Sensitive Net-
performance level identical to dedicated networks. Thus, the works by the Use of Software-Defined Networking,’ in IEEE Interna-
approach lends itself to realizing wireline network slicing. tional Conference on Emerging Technologies and Factory Automation
(ETFA), 2019, pp. 216–223.
Future work will focus on enhancing and extending the [19] F. Bullo, Lectures on Network Systems, 1.6. Kindle Direct Publishing,
setup via integration into a realistic Smart Grid deployment 2022. [Online]. Available: http://motion.me.ucsb.edu/book-lns.
scenario in scope of a cellular energy grid laboratory demon- [20] G. Stomberg, A. Engelmann and T. Faulwasser, ‘A compendium
of optimization algorithms for distributed linear-quadratic MPC,’ at-
strator. Moreover, ongoing O-RAN and 6G research will Automatisierungstechnik, vol. 70, no. 4, pp. 317–330, 2022.
be strengthened by combining our approach with machine [21] ——, ‘Decentralized non-convex optimization via bi-level SQP and
learning to continuously improve parametrization. Here, 6G ADMM,’ arXiv preprint:2204.08786, 2022.
[22] F. Kurtz, I. Laukhin, C. Bektas and C. Wietfeld, ‘Evaluating Software-
energy efficiency will also be studied in depth. Defined Networking-Driven Edge Clouds for 5G Critical Communica-
tions,’ in International Conference on Information and Communication
ACKNOWLEDGEMENT Technology Convergence (ICTC), Jeju, Korea, 2018, pp. 405–410.
This work has been partly funded by the Federal Ministry for Economic [23] Floodlight SDN OpenFlow Controller Version 1.2, 2022. [Online].
Affairs and Climate Action (BMWK) via the project 5Gain under funding ref- Available: https://github.com/floodlight/floodlight.
erence 03EI6018C and is supported by the Federal Ministry of Education and [24] F. Kurtz, R. Wiebusch, D. Overbeck and C. Wietfeld, ‘Predictive 5G
Research (BMBF) via the project 6GEM (funding reference 16KISK038). The Uplink Slicing for Blockchain-driven Smart Energy Contracts,’ in IEEE
authors thank Henrik Ebel for his contributions to the MPC implementation. Conference on Communications Workshops, 2022, pp. 19–24.
[25] N. Dorsch, F. Kurtz and C. Wietfeld, ‘Enabling Hard Service Guar-
R EFERENCES antees in Software-Defined Smart Grid Infrastructures,’ Computer
[1] D. Kumar and N. Pindoriya, ‘A Review on 5G Technological Inter- Networks, vol. 147, pp. 112–131, 2018, Elsevier.
vention in Smart Grid,’ in Power Systems Conference, 2020, pp. 1–6. [26] IEEE Standard for a Precision Clock Synchronization Protocol for
[2] 3GPP, ‘Technical Specification Group Services and System Aspects; Networked Measurement and Control Systems, IEEE Std 1588, 2019.
System architecture for the 5G System (5GS); Stage 2; Release 17,’ 3rd [27] D. Burk, A. Völz and K. Graichen, ‘A Modular Framework for
Generation Partnership Project (3GPP), Technical Specification (TS) Distributed Model Predictive Control of Nonlinear Continuous-Time
23.501, 2022. Systems (GRAMPC-D),’ Optimization and Engineering, vol. 23, no. 2,
[3] X. Foukas, G. Patounas, A. Elmokashfi and M. K. Marina, ‘Net- pp. 771–795, 2022.
work Slicing in 5G: Survey and Challenges,’ IEEE Communications [28] H. J. Ferreau, C. Kirches, A. Potschka, H. G. Bock and M. Diehl,
Magazine, vol. 55, no. 5, pp. 94–100, 2017. ‘qpOASES: A parametric active-set algorithm for quadratic pro-
[4] A. Nota, S. Saidi, D. Overbeck, F. Kurtz and C. Wietfeld, ‘Providing gramming,’ Mathematical Programming Computation, vol. 6, no. 4,
Response Times Guarantees for Mixed-Criticality Network Slicing in pp. 327–363, 2014.
5G,’ in Design, Automation & Test in Europe Conference & Exhibition [29] J. A. Andersson, J. Gillis, G. Horn, J. B. Rawlings and M. Diehl,
(DATE), 2022, pp. 552–555. ‘CasADi: a software framework for nonlinear optimization and optimal
[5] O-RAN ALLIANCE Specifications, 2022. [Online]. Available: https : control,’ Mathematical Programming Computation, vol. 11, no. 1,
//orandownloadsweb.azurewebsites.net/specifications. pp. 1–36, 2019.
[6] N. H. Mahmood, S. Böcker, I. Moerman, O. A. López, A. Munari, [30] D. Moore, E. Olson and A. Huang, ‘Lightweight Communications
K. Mikhaylov, F. Clazzer, H. Bartz, O.-S. Park, E. Mercier et al., and Marshalling for Low-Latency Interprocess Communication,’ Mas-
‘Machine Type Communications: Key Drivers and Enablers towards sachusetts Institute of Technology - Computer Science and Artificial
the 6G Era,’ EURASIP Journal on Wireless Communications and Intelligence Laboratory, Technical Specification (TS) MIT-CSAIL-TR-
Networking, vol. 2021, no. 1, 2021. 2009-041, 2009.
[7] IEEE Standard for Ethernet, IEEE Standard 802.3, 2015.