Time-Sensitive Networking (TSN) is currently being developed at the IEEE as a novel technology that offers an entirely new level of determinism in standard IEEE 802.1 and IEEE 802.3 Ethernet networks. This means that future Ethernet networks will be able to provide:
- Highly limited latency fluctuations (jitter).
- Low packet loss.
- Calculable, guaranteed end-to-end latencies.
- However, to which applications are these characteristics really relevant and how does TSN achieve this functionality?
Today, latency guarantees are established as a basic requirement for real-time data transmissions in a number of application scenarios. These include synchronised axles and drives, power generation, transmission and distribution networks, as well as the transportation industry. In these fields, the cycle times for the transmission of time-sensitive process data are often below one millisecond. To achieve these low cycle times, with correspondingly low latency guarantees, real-time communication technologies such as EtherCAT, PROFINET IRT or SERCOS III are being used. These technologies commonly incorporate additional mechanisms to provide latency guarantees that, in turn, are often incompatible with each other.
As a result, the real-time Ethernet solution market is fragmented and, due to the lack of compatibility, is crippled with regards to future development. TSN has the potential to open up the real-time Ethernet market by establishing a universal physical and data-link layer that is standardised by the IEEE 802. With hard real-time requirements, additional application domains such as process automation can profit from TSN as well. This might seem contradictory because the cycle times in these domains are often larger than, for example, for synchronised drives. For these application scenarios, the benefits of TSN originate in the requirement for guaranteed end-to-end latencies.
In current networks, these guarantees are approximated by over-provisioning the available bandwidth. In contrast, with TSN, it is possible to eliminate such approximation-based solutions and to tailor both the guaranteed bandwidth as well as the latency exactly to the application requirements. Consequently, TSN permits you to plan and to dimension future automation networks according to bandwidth requirements.
When looking at the future of automation networks, a consistent increase in the significance of TSN is foreseeable. Even today, the field of industrial automation is in a period of transition that is driven by the vision of permitting more flexible, intelligent and dynamic production facilities than is currently possible. Terms that are often associated with this vision are Industry 4.0 and Industrial Internet of Things (IIoT). They describe intelligent production environments in which production machinery, conveyor systems and workpieces are constantly communicating with each other in order to support an automated and more efficient production process. This is made possible by increased networking of the sensors and actuators that are involved in the production processes.
Another factor is the increased integration of the (local) cloud, where virtual PLCs are hosted and interact directly with the production process through the sensors and actuators at the field level. These changes affect the models on which the development and planning of current automation networks are based. The familiar automation pyramid is expected to transform into an automation pillar in a long-term, continual process. In contrast to the automation pyramid, where real-time requirements for data transmissions were mostly present at the field level, both the field and the connectivity level will need to fulfill low-latency requirements in the case of the automation pillar.
Another paradigm is emerging beyond the requirements for calculable and lowest-possible latency and jitter – the increased convergence of the different networks that are still used in parallel within existing production sites. While in current facilities, time-sensitive control data is often transmitted via dedicated networks built only for a particular purpose, it is foreseeable that in the future this control data will be transmitted in parallel with “Best Effort” data (e.g. configuration and monitoring data) and data with “soft” real-time requirements (e.g. video data from surveillance cameras) over a common network infrastructure. One key characteristic of TSN is to offer a solution for such converging network infrastructures with high demands on bandwidth at the connectivity level and hard as well as soft real-time requirements at the field and connectivity levels.
TSN adds a level of determinism to Ethernet-based data communication that is able to meet high demands of modern control networks. Even today, it is foreseeable that TSN will reach a broad audience and the target markets of TSN will likely differ from one another significantly. Therefore deterministic, as well as fault-tolerant data transmissions, may be a firm requirement in one target market, while in another case, fault-tolerance through redundant transmissions may only be of secondary importance.
Hence, TSN has been conceived as a modular system by which the precise characteristics of the implementation − and the associated hardware and software requirements − can be tailored to fit the individual requirements.TSN is not made up of a single standard document, but is a family of standards that have been in development by the IEEE 802.1 TSN Task Group since 2012. By now, these activities have yielded their first results – central mechanisms of the TSN family are already available as standard documents.
Until now, it was not possible with Class of Service (CoS) mechanisms such as the IEEE 802.1Q strict priorities to guarantee bounded end-to-end latency of time-sensitive data traffic. Due to queueing effects, an Ethernet frame with low priority that is already in transmission could delay Ethernet frames of even the highest priority at every Ethernet switch along the transmission path. As one of the central components of TSN, the Time-Aware Scheduler (TAS), for the first time, introduces the possibility for prioritising the data transmission of conventional Ethernet frames based on transmission time and guaranteeing their forwarding and delivery at a defined point in time.
The fundamental idea of this TSN mechanism, published as Standard IEEE 802.1Qbv-2016 in March 2016, is to utilise TDMA (Time Division Multiple Access) to divide time into discreet segments of equal length, so-called cycles. This allows dedicated time slots to be provided for the transmission of data packets with real-time requirements within the cycles.
With the aid of the TAS, the transmission of conventional Best Effort Ethernet traffic can be temporarily interrupted in order to forward time-sensitive data traffic within the reserved time slots for high-priority traffic. The TAS thus permits the prioritisation of periodic real-time data in relation to conventional Best Effort data traffic.
Similar to the strict prioritisation scheme, the TAS uses the CoS priorities (PCP – Priority Code Point) that are present in the VLAN tag of the Ethernet header. In this case, all Ethernet frames are processed until they reach the Time-Aware gate queues at the output port. At this point, the TAS intervenes in the packet processing. More precisely, with the use of the TAS, the selection of the next Ethernet frame to be transmitted is no longer just determined strictly by a linear hierarchy at the queue, but rather the state of the respective gates is also taken into consideration. This state may be either open or closed, based on actual time. Ethernet frames that are waiting for transmission in the associated queues will be considered in the packet selection, depending on these states.
For example, only the queue with a priority of seven is processed at this particular point in time. The Gate Control List determines which traffic queue is permitted to transmit at a point in time within the cycle. Besides the states of the Time-Aware Gates, the Gate Control List indicates the length of time for which a specific entry will be active. In the case of the Gate Control List, the list mirrors the cycle that consists of a Best Effort phase, as well as a phase with prioritised data traffic.
Due to the poor predictability of Best Effort traffic patterns, it is generally not foreseeable when a specific Best Effort data packet will need to be processed. For example, the transmission of an Ethernet frame in time slot 2 could be initiated too late. This Ethernet frame would then, despite the use of the TAS, extend into the time slot number 1 of the subsequent cycle. This would therefore result in a delayed processing of real-time data and a violation of guaranteed end-to-end latencies.
In order to avoid these situations, besides the transmission barriers between the time slots, the so-called guard bands have to be introduced in conjunction with the TAS. These guard bands suppress the transmission of packets for the duration of a maximum-size Ethernet frame. Thus, the guard bands can prevent the transmission of Best Effort Ethernet frames that would intrude into the subsequent time slot preventing delays in processing of real-time data during the transition from a Best Effort phase to a phase with high-priority traffic. But this guard band also inevitably results in undesirable dead times where the network can’t be utilised at all and is a waste of bandwidth.
In addition to the configured guard bands, the TAS also permits that the packet length of the next-in-line Ethernet frame is taken into account. The decision whether to transmit now or wait for the next Best Effort timeslot depends on whether the next frame is short enough to be fully transmitted within the current time slot. But even with this mechanism, situations can occur where there is simply not enough time left in the current timeslot, or the frame to be transmitted is too large to fit in the packet. Therefore, even with this mechanism, the dead times that result from the guard bands cannot be entirely prevented.
In order to maximise the usable bandwidth for Best Effort Ethernet frames, the IEEE 802 working group developed a method for Ethernet frame pre-emption completed in June 2016. With this method, conventional Ethernet frames can be divided into partial packets (framelets) of as small as 64 bytes, and each framelet may be transmitted separately. This permits starting the transmission of a large Ethernet frame, despite insufficient remaining time within the Best Effort phase. The frame can be interrupted at the last 64-byte boundary before the current time slot ends and can then be completed in the next Best Effort phase. Frame pre-emption makes it possible to reduce the guard band to the maximum size of one Ethernet framelet.
In the case of a fast Ethernet network the dead time from each guard band can be reduced to 0.12 ms, which means an improvement of the use of the bandwidth available can be achieved. Due to the fact that frame pre-emption is an intrusion into the normal process of Ethernet frame forwarding and processing, it is necessary for both devices of an Ethernet connection (e.g. two Ethernet switches) to announce their support for this mechanism through the use of the Link Layer Discovery Protocol (LLDP – IEEE 802.1AB-20165).
Only with frame pre-emption support on both ends of the link, the feature can be activated on the corresponding end devices or switch ports. With this, backwards compatibility with existing Ethernet devices is maintained.
The TAS utilises only local configuration data – the data that is available in a particular network device (end device or switch). For example, this configuration data consists of information about the lengths of cycles and time slots. Therefore, besides the TAS, close coordination between the devices in the network is required in order to ensure that the frames match the correct time slots in each switch. This enables the transmission of communication streams that can be transmitted through end-to-end connections, with guaranteed latencies and without queuing times. This means, in particular, that all network participants must possess a common understanding of time.
In particular, all participants must know when a cycle begins and which time slot is active in the cycle. In order to enable this, the use of a protocol for time synchronisation, such as the Precision Time Protocol (PTP) in accordance with IEEE 1588 (IEEE 1588-20086) or the IEEE 1588 Profile IEEE 802.1AS (IEEE 802.1AS-20117) is mandatory.
Both IEEE 1588 as well as IEEE 802.1AS permit the synchronisation of distributed clocks within a network with an accuracy of under 1μs. Implemented in hardware, timing precision in the range of a few nanoseconds can be achieved.
With the method of Ethernet frame pre-emption, the guard band size can be reduced from the maximum size of an Ethernet frame to the size of a partial packet.