Table Of Contents
Table Of Contents

Time-Sensitive Networking

This chapter describes the part of INET Framework that implements a subset of the IEEE standards related to Time-Sensitive Networking (TSN). The supported TSN features include among others: time synchronization, per-stream filtering and policing, scheduling and traffic shaping, frame replication and elimination, frame preemption, cut-through switching, automatic network configuration for failure protection, stream redundancy, and gate scheduling.

The above TSN features are implemented in several modules separate from other standard Ethernet functionality. Some of the modules presented in this chapter are specific to the IEEE standards, but several of them are more generic and they are simply reused to achieve the required functionality.

However, it must be noted that several TSN features are only partially supported. The following sections describe the individual TSN features in detail.

Devices

TSN networks require additional functionality compared to standard Ethernet. To facilitate the configuration of the TSN features, INET provides several TSN specific network devices. These are derived from the basic INET network nodes and they provide several additional TSN specific parameters. The additional parameters mostly determine the internal module structure of the TSN specific network nodes, and also further parameterize the basic network node submodules. The usage of these network nodes is completely optional, it is possible to combine the TSN modules in other ways. However, it is advisable to start making TSN simulations using them.

  • TsnClock models a hardware device that exclusively acts as a gPTP master node for time synchronization

  • TsnDevice models a TSN specific end device capable of running multiple applications and all other supported TSN features

  • TsnSwitch models an Ethernet switch capable of all supported TSN features

Using the TSN specific network nodes and having the TSN specific features enabled is just the first step for using Time-Sensitve Networking. Most TSN features require additional configuration in the corresponding modules.

Time Synchronization

This section describes the modules that implement a subset of the IEEE 802.1AS standard titled as Timing and Synchronization for Time-Sensitive Applications (IEEE 802.1AS-2020). There are two main components for this TSN feature, clock modules that keep track of time in the individual network nodes, and time synchronization protocol modules that synchronize these clocks. All required modules are already included in the TSN specific network nodes.

If the default TSN specific network nodes are not used, then the following modules can still be used to keep track of time in the network nodes.

Similarly to the above the following gPTP time synchronization related protocol modules and network nodes can also be used to build time synchronization in a network:

  • Gptp the gPTP time synchronization protocol

  • GptpBridge models a gPTP time synchronization bridge network node

  • GptpEndstation models a gPTP time synchronization end station network node

  • GptpMaster models a gPTP time synchronization master network node

  • GptpSlave models a gPTP time synchronization slave network node

In order to implement node failure (e.g. master clock) and link failure (e.g. between gPTP bridges) protection, multiple time synchronization domains are required. These time domains operate independently of each other and it’s up to the clock user modules of each network node to decide which clock they are using. Typically they use the active clock of the MultiClock and there has to be some means of changing the active clocks when failover happens. The following modules can be used to implement multiple time domains:

  • MultiClock contains several subclocks for the different time domains

  • MultiDomainGptp contains several gPTP submodules for the different time domains

The following parameters can be used to enable the gPTP time synchronization in various predefined network nodes:

  • hasTimeSynchronization parameter enables time synchronization in TSN specific network nodes

  • hasGptp parameter enables the gPTP time synchronization protocol in gPTP specific network nodes

Per-stream Filtering and Policing

This section describes the modules that implement a subset of the functionality of the IEEE 802.1Q standard that was originally introduced by the Per-Stream Filtering and Policing (IEEE 802.1Qci-2017) amendment.

The simplest module for IEEE 802.1Q per-stream filtering and policing is the SimpleIeee8021qFilter compound module. This module combines several submodules: a packet classifier at the input, a packet multiplexer at the output, and one packet meter, one packet filter, and one packet gate per stream. Each one of the latter per-stream 3 modules are optional.

When a packet arrives at the input of the SimpleIeee8021qFilter, it first gets classified into one of the filtering and policing submodule paths. Then the packet meter measures the packet as part of the packet stream that was seen so far, and attaches the result of the measurement. The result may be as simple as a label on the packet. After the metering, the packet filter checks if the packet matches the required conditions and either lets the packet go through or drops it. Finally, the packet gate allows the automatic time based or programmatic control of the packet passing through the selected path of the policing module. Packets are never enqueued in the SimpleIeee8021qFilter, they either pass through or get dropped immediately.

Note that any of the SimpleIeee8021qFilter default submodules can be replaced with other variants. Moreover, other more complicated internal structures are also possible, this is especially the case when the packet meters are replaced with token bucket classifiers as described below.

As the first step, the default policing process starts with a packet classifier, module, the StreamClassifier by default, that classifies packets based on the attached stream information. This classifier simply maps stream names to output gate indices. Please note that the stream decoding and identification process is not part of the SimpleIeee8021qFilter.

In the second step, the default policing process continues with a packet meter module, the DualRateThreeColorMeter by default, that labels the packets either as green, yellow or red based on the committed and excess information rate, and the committed and excess burst size parameters.

The most commonly used packet meters for per-stream filtering and policing are:

The above modules are based on the following generic token bucket meter modules:

  • TokenBucketMeter contains a single token bucket and labels packets one of 2 labels

  • MultiTokenBucketMeter contains an overflowing chain of N token buckets and labels packets with one of N+1 labels

Different packet meter modules can also be used by replacing the default packet meter submodules of SimpleIeee8021qFilter. See the inet.queueing.meter NED package for alternatives.

In the third step, the default per-stream filtering and policing process continues with a packet filter module, the LabelFilter by default, that drops the red packets and lets through the green and yellow ones by default. Of course, different packet filter modules can also be used by replacing the default filter submodules of SimpleIeee8021qFilter. See the inet.queueing.filter NED package for alternatives.

Finally, the default policing process finishes by merging the per-stream filtering and policing paths into a single output gate by using the generic PacketMultiplexer module. There’s no need to prioritize between the per-stream paths here, because the packets pass through in zero simulation time.

Different per-stream filtering and policing compound modules can also be created by combining the existing queueing and protocol element modules of the INET Framework. For example, instead of the packet meter modules, the token bucket based packet classifier modules give more freedom in terms of the module structure. See the inet.queueing NED package for more modules.

The most commonly used packet classifiers for per-stream filtering and policing are:

The above modules are derived from the generic token bucket classifier modules. These modules can also be used on their own and combined in many different ways with all the other queueing modules to achieve the desired per-stream filtering and policing.

There is also a more complex per-stream filtering and policing module, called the Ieee8021qFilter. This module is more similar to the architecture that is present in the IEEE 802.1Q standard. The Ieee8021qFilter also combines several submodules but in a slightly different way than the SimpleIeee8021qFilter. The most important difference is that this module can be mostly configured through a single streamFilterTable parameter.

The TSN specific network node TsnDevice and TsnSwitch have a special parameter called hasIngressTrafficFiltering which can be used to enable the traffic filtering and policing in the network node architecture. Of course, these modules can also be used in other ways.

Scheduling and Traffic Shaping

This section describes the modules that implement a subset of the functionality of the IEEE 802.1Q standard that was originally introduced by the Enhancements for Scheduled Traffic (IEEE 802.1Qbv-2015) amendment.

The traffic shaping architecture is part of the queue submodule of the MAC layer in the network interface. Currently three different packet shaper algorithms are supported, the credit-based shaper, the time-aware shaper, and the asynchronous shaper. In order to configure the network interface to use traffic shaping the queue submodule must be replaced with either the GatingPriorityQueue or the PriorityShaper compound modules. Both contain a packet classifier to differentiate between the traffic categories and a priority packet scheduler that prefers higher priority traffic categories over lower priority ones. The difference is in the structure of the other submodules that form the shapers.

The credit-based shaper is implemented in the CreditBasedShaper module using a standard PacketQueue and a special purpose Ieee8021qCreditBasedGate submodule. The latter module keeps track of the available credits for the given traffic category and allows or forbids the transmission of packets.

The time-aware shaper is implemented in the TimeAwareShaper compound module that uses a standard PacketQueue and a special purpose PeriodicGate. The latter module has parameters to control the gate schedule that determines the periodic open and gate.

The asynchronous shaper is in part implemented in the AsynchronousShaper compound module. This shaper is somewhat more complicated than the previous two because it also contains submodules that are part of the ingress per-stream filtering module in the bridging layer. These are the EligibilityTimeMeter and the corresponding EligibilityTimeFilter submodules. The first is responsible for calculating the transmission eligibility time for incoming packets, the latter is responsible for dropping packets which are considered to be too old for transmission. The shaper in the network interface queue contains two additional submodules called EligibilityTimeQueue and EligibilityTimeGate. The former is responsible for sorting the frames according to the transmission eligibility time, the latter is a gate that is open only if the transmission eligibility time of the first frame of the queue is greater than the current time.

The TSN specific network node TsnDevice and TsnSwitch have a special parameter called hasEgressTrafficShaping which can be used to enable the traffic shaping in the network node architecture. Of course, these modules can also be used in other ways.

Frame Replication and Elimination

This section describes the modules that implement a subset of the functionality of the IEEE 802.1CB standard titled as Frame Replication and Elimination for Reliability (IEEE 802.1CB-2017).

The relevant modules are all part of the BridgingLayer compound module that resides between the network layer and link layer protocols. This compound module also contains other functionality such as frame forwarding. There are four relevant submodules, each one implements a very specific part of frame replication.

The first part deals with stream identification, and is implemented in the StreamIdentifierLayer module and its StreamIdentifier submodule. This module is only useful in network nodes which produce application traffic themselves. The stream identifier module is responsible for assigning a stream name for outgoing packets by looking at their contents and meta data. For example, packets can be identified by the destination MAC address and PCP request tags. Since at this point the packets don’t yet contain any layer 2 header the decision can be based on the attached request tags that will be later turned into packet headers.

The second layer handles incoming stream merging and outgoing stream splitting. This layer is called the StreamRelayLayer and contains two submodules called StreamMerger and StreamSplitter. The former is responsible for merging incoming member streams into a single stream and removing duplicate frames. The latter is responsible for splitting outgoing streams into potentially several member streams.

The third part deals with ingress and egress stream filtering, and it’s implemented in the StreamFilterLayer module that contains one submodule for both directions. This part is not strictly necessary for frame replication. Most often only the ingress filtering submodule is used as described in the previous section.

The last layer handles incoming packet decoding and outgoing packet encoding. This module is called the StreamCoderLayer and it contains two submodules the StreamDecoder and StreamEncoder. The former handles the stream decoding of incoming packets by checking the attached indication tags. The latter deals with the encoding of outgoing packets by attaching the necessary request tags.

The TSN specific network node TsnDevice and TsnSwitch have a special parameter called hasStreamRedundancy which can be used to enable frame replication in the network node architecture. Of course, these modules can also be used in other ways.

Frame Preemption

This section describes the modules that implement a subset of the functionality of the IEEE 802.1Q standard that was originally introduced by the Frame Preemption (IEEE 802.1Qbu) amendment.

Frame preemption requires the network interface to be able to interrupt an ongoing transmission and switch to the transmission of a higher priority frame. This behavior is implemented in special MAC and PHY layer modules that use packet streaming in the network interface. This is in contrast with the default behavior where modules pass packets around as a whole.

The TSN specific network nodes, TsnDevice and TsnSwitch, have a special parameter called the hasFramePreemption which can be used to enable frame preemption in the network interfaces. Of course, these modules can also be used in other ways.

Cut-through Switching

The default store and forward mechanism in Ethernet switches greatly influences the end-to-end latency of application traffic. This effect can be overcome and drastically reduced by using cut-through switching. This methods starts forwarding the incoming frame before the whole frame has been received, usually right after the reception of the MAC header. However, cut-through switching is not a standard mechanism and there are all kinds of variants in operation.

INET provides the following modules related to cut-through switching:

  • EthernetCutthroughInterface models an Ethernet interface that contains a special cut-through layer between the MAC and PHY layers that in certain circumstances allows the direct forwarding of frames from the incoming network interface to the outgoing

  • EthernetCutthroughLayer models the cut-through layer with direct connections to other cut-through interfaces inside the same network node

  • EthernetCutthroughSource models the source of the cut-through forwarding inside the network interface

  • EthernetCutthroughSink models the sink of the cut-through forwarding inside the network interface

Surprisingly cut-through switch also has to be enabled in the end devices because the receiving switch has to be notified both at the start and at the end of the frame.

The TSN specific network nodes, TsnDevice and TsnSwitch, have a special parameter called the hasCutthroughSwitching which can be used to enable cut-through switching in the network interfaces. Of course, these modules can also be used in other ways.

Automatic Network Configuration

Configuring the features of Time-Sensitive Networking in a complex network that contains many applications with different traffic requirements is a difficult and error prone task. To facilitate this task, INET provides three types of network level configurators:

  • gate scheduling configurators are capable of configuring the gate control lists (i.e. periodic open/close states) for all traffic classes in all network interfaces based on packet length, packet interval, and maximum latency parameters

  • stream redundancy configurators are capable of configuring the stream merging and stream splitting modules as well as the stream identification in all network nodes to form the desired redundant streams for each application traffic

  • failure protection configurators are capable of using the previous two to achieve the desired link and node failure protections for all streams in the network based on the set of failure cases

All other network level configurators such as the Ipv4NetworkConfigurator or the MacForwardingTableConfigurator can also be used.

There are several different automatic gate scheduling configurators having different capabilities:

There is only one stream redundancy configurator:

Currently there is only one failure protection configurator:

  • FailureProtectionConfigurator configures the gate scheduling and the stream redundancy configurators to provide protection against the specified link and node failures

All of these configurators automatically discover the network topology and then taking into account their own independent configuration they compute the necessary parameters for the individual underlying modules and configure them. However, anything they can do, can also be done from INI files manually, and the result can also be seen at the configured module parameters in the runtime user interface.