Time-sensitive networking for fast, reliable networks

It is challenging to predict Ethernet and IP network delays for packets traversing the network. Possible loss and delay variations make Ethernet unsuitable for a variety of applications, including automotive, industrial control processes, 5G fronthaul transport, and IoT, which demand low, deterministic latency and reliable message delivery.

IEEE Time-Sensitive Networking (TSN) consists of extensions to existing mechanisms for forwarding traffic through Ethernet switches, adding determinism, low transmission delay, and reliability. Applications may adopt some or all TSN recommendations to achieve the desired performance and cost.

Traffic scheduling and shaping

Applications differ in their need for bandwidth, delay, jitter, and loss tolerance. For example, surveillance camera video requires high bandwidth but is tolerant of loss and jitter. On the other hand, automated production control signals require low bandwidth but cannot tolerate loss or delay. So, a large video frame transmitted through a switch makes smaller process-control frames wait in a queue for forwarding, making the network unsuitable for process control. IEEE 802.1Qbv time-aware scheduling is recommended to address this challenge.

802.1Q VLAN tagged frames can have one of eight priorities. Traditional switches provide differential treatment based on priority. They either allocate more bandwidth based on priority – for example, video can get more bandwidth – or forward frames with one priority tag before frames with another priority tag. Unfortunately, these are insufficient to ensure that our process control signal is always forwarded within a fixed upper delay limit.

802.1Qbv proposes a TDM scheme with time slots. Each priority tag can be assigned to a specific time slot. If assigned an exclusive time slot, our process control signal can be forwarded with a fixed upper bound on the delay. (See Figure 1.)

Scheduling without 802.1Qbv

Time-aware scheduling with 802.1Qbv

Figure 1. Scheduling without 802.1Qbv and time-aware scheduling with 802.1Qbv
Source: Capgemini Engineering

802.1Qav places a requirement on endpoints and switches to implement traffic shaping. Devices implementing shaping transmit frames at an even pace, avoiding bursts. This approach smooths out network utilization and reduces variations in forwarding delay.

Frame pre-emption

The IEEE 802.1Qbu frame pre-emption recommendation allows switches to break up a large frame at one end of a link and reassemble the fragments at the other end to transmit other high-priority frames between the fragments. Switches advertise this capability and learn their neighbor switch’s capability through link layer discovery protocol (LLDP) messages since both sender and receiver must implement this at the link level. After starting transmission of a large frame, a switch can suspend transmission of the frame. Then, the switch can transmit delay-intolerant time-sensitive frames and then resume the transmission of the large frame. The receiving switch stores the fragments and reassembles them to re-form the original frame. (See Figure 2.)

Without frame pre-emption – small frames wait until a large frame is transmitted

Figure 2. With and without 802.1Qbu frame pre-emption
Source: Capgemini Engineering

Time synchronization

All networked devices must synchronize their clocks to measure and monitor transmission delay and loss, thereby minimizing latency and loss. However, installing a GPS clock in every device may be logistically impossible, such as in coal mines and mountain tunnels, and cost-prohibitive, such as in sensors. The IEEE 1588 Precision Time Protocol (PTP) is a way for network elements to have a common understanding of time. (See Figure 3.)

In this scheme, a grandmaster clock in a network obtains its clock from a high-precision source such as GPS. The grandmaster transmits time synchronization information in Ethernet frames to all other devices. Endpoint devices, such as robots, use the information to synchronize their clocks continuously with the grandmaster. Network elements between the grandmaster and endpoint devices have roles such as master, boundary, or transparent lock.

PTP allows a large number of options for different scenarios. IEEE 802.1AS defines a specific choice of options for industrial and automotive applications.

Network with time synchronized elements; the transparent clock and boundary clock are usually implemented in network elements like switches and routers

Figure 3. IEEE 1588 PTP for network elements to
synchronize their clocks
Source: Capgemini Engineering

Stream reservation protocol

Traffic scheduling and shaping are effective only when the network has sufficient resources. IEEE 802.1Qat stream reservation protocol (SRP) addresses resource sufficiency for specific application traffic. The source of a stream/flow sends reservation messages through the switches along the intended traffic path, which then checks the availability of resources — for example, bandwidth in links and buffers in queues — to support the traffic stream and confirm or deny the reservation. This approach ensures network resource availability for time-sensitive and critical application traffic. (See Figure 4 .)

Resource reservation confirmation

Figure 4. 802.1Qat resource reservation for time-sensitive traffic
Source: Capgemini Engineering

The IEEE 802.1Qcc recommendation improves on this approach to meet automotive, consumer, and industrial communications requirements by adding support for configurable reservation classes, enhancing descriptions for stream characteristics, providing a larger number of streams and support for Layer 3 streaming, and a user-network-interface (UNI) for routing and reservations.

Path control, selection, and reservation 802.1QCA

For the SRP to reserve resources at each network element in a path, the traffic source, or the switch at the head end, must know the route to the destination. This is achieved by an extension to the IS-IS routing protocol that allows Ethernet switches or end-stations to advertise their connectivity and learn the network topology through advertisement messages from other switches or nodes.

Frame replication and elimination for reliability

IEEE 802.1CB describes a mechanism to achieve redundancy in a network that is susceptible to packet loss. Messages are copied and communicated in parallel over disjointed paths through the network. At the receiver end, redundant duplicates are eliminated to create one seamless stream of information. This method ensures high availability for data sent over TSN networks. It is similar to high-availability seamless redundancy (HSR) and parallel redundancy protocol (PRP) defined in International Electrotechnical Commission (IEC) 62439.

Redundancy is achieved by selecting and ensuring more than one disjointed path for the traffic to be sent over from the source. Redundant paths are created by the stream reservation and path control mechanisms described previously.

Summary

5G, IoT, and Industry 4.0 use cases demand high determinism and reliability from Ethernet networks. Time-Sensitive Networking from the IEEE is the answer that enables the benefits of cost and simplicity provided by Ethernet to these new networking use cases.


Author

Valérie Sauterey

Rajesh Kumar Sundararajan

Consultant, Capgemini Engineering

Rajesh has 25 years of experience in the datacom and telecom industry spanning engineering, marketing, quality control, product management, and business development. He is always connected to the technology and has been involved in projects in IP, routing, MPLS, Ethernet, network access, network aggregation, transport networking, industrial networking, data-center networking, network virtualization, and SDN technologies.

Contact US Capgemini Engineering

MEET OUR EXPERTS

You can work with a company built for now,
or you can with one engineering the virtual networking software of tomorrow.

GET IN TOUCH