Software-defined Networking for Real-time Ethernet
Jia Lei Du and Matthias Herlich
Salzburg Research, Jakob Haringer Straße 5/3, Salzburg, Austria
Keywords: Software-defined Networking, Real-time Ethernet.
Abstract: Real-time Ethernet is used in many industrial and embedded systems, but has so far mostly been statically
configured. However, in the future these network configurations will be required to change dynamically, for
example for highly flexible production lines or even software upgrades in modern cars that add new features
which require changes to the in-vehicle network. Software-defined networking (SDN) is already
increasingly used to dynamically configure non-real-time networks. In this paper we explore the idea of a
software-defined real-time Ethernet. We analyze the features of current real-time Ethernet protocols, the
applicability of SDN and give an overview of potential advantages of software-defined networking for real-
time communication which can enable features not achievable using current solutions. In the future this
development will likely lead to more flexible, efficient and robust real-time networks.
1 INTRODUCTION
Real-time Ethernet (RTE) allows the use of cost
effective, widespread and high-bandwidth Ethernet
technology in industrial environments like
automation, process control and transportation
where one key challenge is real-time
communication, i.e. communication with guaranteed
upper bounds for latency and latency variations
(jitter). Various solutions like Ethernet Powerlink,
VARAN, Profinet and TTEthernet have been
developed to extend standard Ethernet with real-time
capabilities.
Typical RTE deployments in the past have been
configured once to run without re-configuration for
years or even decades. However, in the future RTE
networks will need to be more flexible due to a
variety of reasons: To produce small lot sizes in a
production environment efficiently, the underlying
network must support quick reconfigurations to
fulfill new requirements (Dürkop, Jasperneite and
Fay, 2015). Or in-vehicle networks could be
reconfigured through software updates for example
when a new driver assistance feature needs a higher
sample rate from a proximity sensor.
In non-real-time networks software-defined
networking (SDN) is a technology that provides a
great range of freedom to flexibly and centrally
reconfigure the network on-demand. The basic idea
of SDN is to control network flows through a
centralized intelligent controller with “dumb”
forwarding devices in the data plane of the network
(McKeown et al., 2008). By monitoring network-
wide state, the controller obtains an up-to-date view
of the network and can dynamically adapt flows as
necessary. The concept of SDN allows a wide range
of traffic engineering, security and other
applications. For example, flows can be dynamically
rerouted based on load, failure or security scenarios
to provide certain bandwidth or latency properties,
fast failover mechanisms or security services. From
an economic point of view, through standardization
and centralization, SDN has the potential to simplify
and reduce costs for network setup and operation.
In this paper we will describe our idea to apply
software-defined networking in real-time Ethernet
networks to benefit from SDN advantages while
keeping the deterministic properties of RTE. In
detail we propose replacing the switches/hubs of
real-time Ethernet solutions with SDN-capable
switches. Note that we do not consider replacing the
real-time protocols themselves but to extend RTE
protocols by providing additional features that the
use of SDN controllers and switches make possible.
For this purpose we first describe SDN in the next
section. Then we describe features typical RTE
solutions provide. Finally, we discuss the advantages
and disadvantages of using SDN in an RTE network
and give an approach how to validate these claims.
584
Du, J. and Herlich, M.
Software-defined Networking for Real-time Ethernet.
DOI: 10.5220/0005996605840589
In Proceedings of the 13th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2016) - Volume 2, pages 584-589
ISBN: 978-989-758-198-4
Copyright
c
2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
2 RELATED WORK
Gopalakrishnan (Gopalakrishnan, 2014) and Kalman
(Kalman, 2014) both consider how SDN can be used
in industrial communication networks.
Gopalakrishnan provides a general list of SDN
features and gives some examples how the
advantages could be applied to an IEC 61850-based
network, but only mentions real-time capabilities in
passing. Kalman focuses on hardware abstractions
and the ability to automatically configure networks
using SDN. While both consider the advantages of
SDN, they do not focus on the specific requirements
and advantages SDN can bring to real-time
networks, but on industrial communication networks
in general.
(Dürkop et al., 2015) provides a high-level
concept for the automatic configuration of real-time
Ethernet solutions. Our paper focuses on the
communication aspect in more detail and proposes
using SDN as an approach for network
(re-)configurations. Furthermore, automatic
(re-)configuration is only one of the advantages we
describe in this paper that software-defined
networking can bring to RTE networks.
3 SOFTWARE-DEFINED
NETWORKING BACKGROUND
In most conventional communication networks,
traffic flows are established based on forwarding
rules that are locally determined using distributed
algorithms. In contrast to this approach, traffic flows
in software-defined networks (SDNs) are centrally
configured by network applications via so-called
controllers. This effectively decouples the control
plane, which determines where traffic is sent, from
the data plane, which forwards packets to their
destinations. When a packet that matches a rule
arrives at a network device, the associated actions
are performed. Possible actions include the
modification of packet headers and the dropping or
forwarding of packets. Figure 1 illustrates the
interaction between lower layer SDN forwarding
devices, the SDN controller with its applications,
and RTE devices.
One standard for the implementation of software
defined networks is OpenFlow (Open Networking
Foundation, 2015). The OpenFlow standard defines
a communication protocol between network
switches and one or more controllers. The ideas in
this paper can be applied to all SDNs, but we will
use OpenFlow as example when illustrating our
ideas.
Figure 1: In SDN the network devices (middle) forward
network flows programmed by a SDN controller (top)
between end-devices (left/right). The SDN controller
could also be integrated into one of the end-devices.
One key risk of an SDN is related to the
availability of the controller that is required for
configuring the network devices. Both the controller
itself and the connection between network devices
and controller represent possible single points of
failures and bottlenecks. To mitigate risks of
controller unavailability usually the use of multiple
controllers in an SDN is suggested such as in
(Yeganeh and Ganjali, 2012; Jain et al., 2013; Yazici
et al., 2014).
4 REAL-TIME ETHERNET
FEATURES
We have exemplarily chosen Ethernet Powerlink
(Ethernet Powerlink Standardization Group, 2016),
Profinet (International Electrotechnical Commission,
2014), TTEthernet (SAE Aerospace, 2011),
VARAN (VARAN Bus User Organization, 2016)
and TSN (Time-Sensitive Networking Task Group,
2016) (an upcoming but not yet finalized IEEE
standard and the successor of AVB) for
investigation. Industrial communication protocols
like Ethernet/IP are implemented on application
layer based on TCP/UDP over IP communication
stacks. Protocols in this category are usually highly
compatible and do not require special hardware or
modifications. However, due to the use of the entire
Internet stack cycle times are generally higher than
those achieved by protocols implemented based on
lower communication layers. The studied protocols
are instead implemented directly on top of Ethernet
and achieve significantly lower cycle times. Thus,
the application of OpenFlow and software-defined
networking to real-time networks of the second
Software-defined Networking for Real-time Ethernet
585
category is technologically more challenging and
findings and improvements are more likely to be
transferable to protocols of the first category.
Additionally, our selection of protocols covers both
time triggered and polling-based protocols as well as
protocols that use the entire Ethernet stack or only
parts of the stack. For an analysis of the features of
the studied real-time Ethernet protocols with respect
to SDN we group the features in the following
categories. (1) Performance: quantifiable
measurements about the RTE solutions. (2)
Compatibility: RTE solutions usage of standard
Ethernet features. (3) Features relevant for SDN:
Specifics of RTE protocols that are relevant for
SDN.
The performance of an RTE can be described by
cycle times and data rate. The cycle time is the
duration of one transmission cycle, which is usually
repeated as long as the network is operating. The
cycle time is relevant for applications that need to
transmit small amounts of data often. The data rate is
the maximum achievable rate of data that can be
transmitted over a single link under optimal
circumstances. The data rate is important for
applications that want to transmit large amounts of
data. Both the cycles times and data rates given in
Table 1 are for optimal conditions and are not
necessarily achievable in practice.
RTE protocols are all based on Ethernet, but use
different network modes and some change Ethernet
standard formats. Network mode describes whether
the RTE currently uses switches or hubs. All RTE
protocols we consider can transmit non-real time
traffic (for example web traffic) in time slots not
reserved for higher priority traffic. VARAN uses its
own kind of frame, while all other protocols we
analyzed use standard Ethernet frames.
RTE solutions have two basic operating
principles: Time scheduled and polling. In polling a
single master server queries all clients according to
its internal schedule. The clients are only allowed to
transmit data in response to a query by the server. In
a time scheduled network a pre-defined schedule is
shared by all devices. The schedule describes which
device is allowed to transmit at which time. While
both time scheduled and polling based RTEs
typically use a schedule, in polling the schedule is
known only to the server and can be changed
dynamically more easily. To use a distributed
schedule precise time synchronization is necessary.
In case of link failures (such as cable breaks)
some RTEs offer redundancy features, which
automatically use alternate links to transmit the data
and thereby hide the failure from the application. A
broadcast (transmission from one-to-all devices) can
be used to implement a multicast (one-to-some) by
filtering out frames at the devices which are not
intended to receive the frame. A more efficient
method which we call real multicast is to transmit
the frame only to the intended receivers in the first
place. Using multipath routing several paths can
deliver data from a source to a destination. This can
be used for redundancy or to increase the data rate.
We define concurrency as the ability of two pairs of
senders and receivers to simultaneously
communicate. This feature is, for example, easily
achieved using switches, but not using hubs.
Network topology describes the configuration of
network devices the RTE solution supports. Hot
plugging is the ability to connect and disconnect
devices during network operation. Note that it is
necessary to prepare the configuration for devices to
be hot plugged in advance in some RTE protocols.
One of SDN’s main capabilities is the fine-
grained control of data flows in the network.
Therefore, RTE features like broadcasting, real
multicasting, concurrency, arbitrary topologies,
redundancy and multipath routing will be as
realizable using SDN as using more traditional
networking approaches – SDN will potentially even
allow a more efficient solution. However, one key
limitation needs to be pointed out: Standard SDN
devices currently do not support frame forwarding at
precise points in time and, thus, do not naturally
support time scheduled protocols. However, adding
the notion of time does not conceptually contradict
the use of SDN and is thus rather an implementation
issue.
5 ADVANTAGES OF
SOFTWARE-DEFINED
REAL-TIME ETHERNET
In this section we discuss the advantages of applying
software-defined networking to the design and
implementation of real-time Ethernet. Some
described features may already be supported or
could be implemented with sufficient effort in
existing solutions. However, even in those cases the
use of SDN would still provide the advantage of
being able to use an existing, consistent framework
to implement all described advantages in a simpler
way. Additionally, SDN enables features (e.g. the
active use of network loops) that are not possible in
existing solutions.
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics
586
5.1 Advantages Not Related to Path
Selection
Central Configuration: Centralized software-based
(re-)configuration of network devices is a key
feature of SDN. It enables centrally controlled
configuration of network nodes both with regard to
device settings and communication patterns (this
advantage has also been named “Flow Engineering”
(Gopalakrishnan, 2014) or “Central Resource
Management” (Kalman, 2014)). In difference to
current RTE solutions where device settings and
communication patterns are often configured once
during design, using an SDN approach device
settings and communication paths and schedules can
be adapted on-the-fly with little or no disruption.
From an application point of view a different
production objective in a factory or a new feature in
an autonomous vehicle could be activated through a
software update even if the requirements towards the
underlying RTE network changes, for example
because certain sensor data is required at a higher
rate or from a different set of connected sensors.
Standardization: First, OpenFlow defines a set of
functionalities that all OF compatible network
devices must fulfill and a standard interface to
access these functions (also mentioned as “open
standards-based and vendor-neutral” in (Kalman,
2014)). Second, as the intelligence is mostly located
in the centralized controller, the network devices are
comparatively simple. These two properties lead to
simple, exchangeable, inexpensive, and future-proof
network devices (except the SDN controller).
Global Network Information: OpenFlow-
compatible network devices can collect a many
usage statistics such as the number of received/sent
frames/bytes per flow/port/queue. This information
can help with error diagnostics and
performance/traffic pattern evaluations. This feature
is more valuable for real-time Ethernet networks in
which the RTE controller does not already have a
comprehensive overview of most or even all
communication.
5.2 Advantages Related to
Switching/Routing/Path Selection
Central Addition and Removal of Network Nodes:
Based on OpenFlow network nodes can be
dynamically added to or removed from the real-time
network at network level and removed nodes would
no longer receive messages. Using this feature
machines, sensors or actuators can, for example, be
dynamically recombined to fulfill different tasks.
Arbitrary Topology: Currently existing protocols
usually support only standard Ethernet topologies
and do not permit the existence of loops on network
level and algorithms like spanning trees protocols
are used to block redundant paths. Due to the central
configuration of communication paths the existence
of loops does not pose a problem for SDN and
arbitrary network topologies can even be actively
exploited.
Fast Reroute and Failover: Additional links in
the network can be used as backup routes in case of
failures in the network. This feature can be more
easily implemented for polling-based RTE protocols
which often use broadcasts. In case of link failures,
frames can be rerouted (Pfeiffenberger et al., 2015)
transparently for end nodes as long as the frames
arrive in time. For time-scheduled protocols the
schedules in the network devices might have to be
Table 1: A comparison of real-time Ethernet protocols and features and their relevance for SDN.
TSN Profinet TTEthernet Powerlink VARAN
Performance
Min. cycle times 30 μs + 31.25 μs <100 μs <100 μs <100 μs
Max. data rate 1 Gbit/s + 100 Mbit/s 1 Gbit/s 100 Mbit/s 100 Mbit/s
Compatibility
Network devices Switches Switches Switches Hubs Hubs
Non-RT traffic Yes Yes Yes Yes Yes
Ethernet frames Yes Yes Yes Yes No
SDN relevant
Operating principle Time schedule Time schedule Time schedule Polling Polling
Redundancy Yes Ring/multi-controller Dual and triple Ring and dual No
Real multicast Yes No ? No No
Broadcast Yes Possible/not used Yes Yes Master to slaves
Multipath routing Yes No Yes No No
Concurrency Yes Yes Yes No No
Topologies Arbitrary Line, tree, star, ring Line, tree, star, ring Line, tree, star, ring Line, tree, star
Hot plugging Yes Yes ? Yes Yes
Software-defined Networking for Real-time Ethernet
587
adapted after an incident to avoid congestion in the
backup paths. For zero-loss/zero-time failover, flows
can be duplicated on the network layer and delivered
via two distinct paths.
Multiple Simultaneous Communication Paths:
Additionally available network links cannot only be
used as backups in case of failures but also to
increase available bandwidth during normal
operation. Even multipath routing is imaginable, that
is, splitting up and delivering flows via multiple
paths.
Multiple Networks over One Infrastructure: An
OpenFlow-/SDN-based approach to RTE networks
could enable or simplify the operation of multiple
real-time Ethernet networks over a single physical
infrastructure, for example, in the most simple case
by reserving half of the time for network 1 and half
of the time for network 2. The devices in the two
networks would never receive messages from the
other network and thus this sharing of the physical
infrastructure could be completely transparent to the
participating devices. Such a setup may require
some form of time synchronization between devices
in the two networks which could take place in a third
virtual network. This feature could be highly
attractive for many polling-based protocols as such
an operation can currently not be supported (due to
the use of broadcasting for communication). For
some time-based protocols like TTEthernet such a
behavior could already be supported conceptually
but the use of SDN would still significantly simplify
the implementation by guaranteeing safety
properties (e.g. nodes in network 1 will never see
messages from nodes in network 2) similar to a
virtualization layer in computing.
Isolation of Faulty Nodes: Using OpenFlow
faulty network nodes can be easily disconnected
from the network in the sense that messages of
faulty nodes can be simply dropped at the closest
functioning network node. The isolation of faulty
network nodes consists of two separate problems:
The detection of faulty behavior through the RTE
and/or SDN controller and the disconnection of the
faulty node through the SDN controller. Detection of
very basic faults can, for example, be done through
simple SDN-based frame counting. For the detection
of complex faults the cooperation between RTE
controller and SDN controller is likely necessary.
Even a selective isolation of a node is possible:
correct frames are allowed to pass and only incorrect
frames that are sent at the wrong time or to wrong
destinations are blocked.
Dynamic Load Balancing: Dynamic load-
balancing allows the dynamic change of communi-
cation paths and/or the simultaneous use of multiple
communication paths between a sender and a
receiver as a function of network load. Within the
scope of this paper/project we use the term only in
the context of asynchronous traffic which potentially
has more volatile communication patterns that are
not known beforehand but less strict latency
requirements compared to isochronous traffic.
Efficient Multicasting: When delivering multicast
traffic using OpenFlow, it is comparatively
straightforward to avoid sending frames over a link
if there is no subscriber of that multicast traffic at
the other end of the link. In difference to standard
Ethernet implementations where multicast frames
are actually broadcasted in the network, this can
both be a security benefit and to save bandwidth.
More efficient bandwidth usage through efficient
multicasting is possible for real-time Ethernet
protocols which allow multiple parallel
communication flows. And protocols that only allow
one sender in the network at any time would still
benefit from a security point of view as nodes that
are not subscribers of the multicast traffic would not
receive any of those frames.
6 DISADVANTAGES AND
POTENTIAL PROBLEMS
Implementation of RTE using current SDN
Technology: The most important feature a RTE
network has to implement is the deterministic
guarantee of traffic latency. To make these
guarantees usually polling or predefined
communication schedules are used. If the creation of
a polling-based software-defined RTE network was
the goal, hubs would have to be replaced with
switches. To implement a software-defined RTE
network based on predefined schedules the SDN
switches would additionally need to have a notion of
time and schedules. While we do not know any
conceptual reason which would prevent the support
of time schedules in SDN switches, we are not
aware of any standard SDN switches which support
schedules. Additionally, when low cycle times are
required, the performance guarantees depend on the
achievable forwarding latency and jitter of SDN
switches. It is necessary to measure the performance
of SDN switches and compare it to current Ethernet
switches and hubs used for RTE. Finally, SDN in
general does not dependent on the use of Ethernet-
compatible frames, however current OpenFlow-
compatible switches do pose that requirement.
ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics
588
Disadvantages Introduced by SDN: One key
disadvantage of SDN is the need for a controller.
Such a controller is a single point of failure (if not
replicated, see section II) and a controller failure
would disable further central network configure-
tions. However, this shortcoming prevents only use
cases in which it is necessary to reconfigure the
network while deterministic traffic is transferred
over the network. In all other cases, the guaranteed
performance would not be affected even if the SDN
controller failed, only reconfiguration would be
disabled.
7 VALIDATION CONCEPT
We are currently developing a proof-of-concept
based on openPowerlink and SDN switches.
openPowerlink is an open source implementation of
the Powerlink real-time Ethernet protocol. A real-
time Ethernet network with a cycle time of 1 ms has
been built based on openPowerlink and OpenFlow-
capable switches in our test lab. We are currently in
the process of implementing key use cases to
demonstrate some of the advantages described in
this paper. Particular emphasis is put on
demonstrating use cases which can be easily
implemented using SDN but would be complex or
impossible to implement using current standard RTE
technologies. Finally, we focused on network level
reconfigurations in this paper. However for the
implementation of some of the described
advantages, a tight integration and interaction with
the respective RTE protocol would be necessary
(e.g. to distribute new time schedules to the network
devices). Thus, the long-term goal is to develop a
complete software-defined real-time Ethernet
solution in which the OpenFlow controller is
integrated in the RTE devices and seamlessly
interacts with the RTE protocols and its features.
8 CONCLUSIONS
We first described software-defined networking and
features of real-time Ethernet solutions from a SDN
point of view. Then we analyzed the advantages and
disadvantages of the application of SDN approaches
to RTE networks and described how we plan to
demonstrate the advantages in practice. We conclude
that the development of a software-defined real-time
Ethernet is a highly promising endeavor and are in
the process of validating our concepts in a test
network.
(This work was partially funded by the Austrian
Federal Ministry for Transport, Innovation and
Technology in the project OpenheaRTEd, FFG No.
849972.)
REFERENCES
Dürkop, L., Jasperneite, J., Fay, A., 2015. An Analysis of
Real-Time Ethernets With Regard to Their Automatic
Configuration. In IEEE World Conference on Factory
Communication Systems (WFCS).
Ethernet Powerlink Standardization Group, 2016. Ethernet
Powerlink Communication Profile Specification.
Version 1.3.0.
Gopalakrishnan, A., 2014. Applications of Software
Defined Networks in Industrial Automation.
International Electrotechnical Commission, 2014.
Additional fieldbus profiles for real-time networks
based on ISO/IEC 8802-3. IEC Standard 61784-
2:2014, section CPF3.
Jain, S., Kumar, A., Mandal, S. et al., 2013. B4:
Experience with a globally-deployed software defined
WAN. ACM SIGCOMM Computer Communication
Review, vol. 43, no. 4, pp. 3-14.
Kalman, G., 2014. Applicability of Software Defined
Networking in industrial Ethernet. In IEEE
Telecommunications Forum (TELFOR).
McKeown, N., Anderson, T., Balakrishnan, H. et al.,
2008. OpenFlow: enabling innovation in campus
networks. ACM SIGCOMM Computer Communication
Review, vol. 38, no. 2, pp. 69-74.
Open Networking Foundation, 2015. OpenFlow Switch
Specification Version 1.5.1.
Pfeiffenberger, T., Du, J. L., Bittencourt, P., et al., 2015.
Reliable and Flexible Com. for Power Systems: Fault-
tolerant Multicast with SDN/OpenFlow. In 7th IFIP
Conf. on New Technologies, Mobility and Security.
SAE Aerospace, 2011. Time-Triggered Ethernet. SAE
Aerospace Standard AS 6802.
Time-Sensitive Networking Task Group, 2016.
http://www.ieee802.org/1/ pages/tsn.html.
VARAN Bus User Organization, 2016. “VARAN Real-
Time Ethernet”.
Yazici, V., Sunay, M. O., Ercan, A. O., 2014. Controlling
a software-defined network via distributed controllers.
arXiv preprint, arXiv:1401.7651.
Yeganeh, S. H., Ganjali, Y., 2012. Kandoo: a framework
for efficient and scalable offloading of control
applications, Workshop on Hot Topics in Software
Defined Networks.
Software-defined Networking for Real-time Ethernet
589