HomePublicationsTechnical Report

On Preemption in Time-Sensitive Networking (TSN)
Ref: CISTER-TR-191202       Publication Date: 29, Nov, 2019

Abstract:
A major requirement to achieve responsiveness, i.e., timely and correct reaction to events, in distributed real-time applications is the ability of data to move on the underlying communication network in a reliable and predictable manner. Ethernet is the emerging communication technology for modern real-time wired networks. Its ability to scale up to the increasingly stringent speed requirements and distance, together with its high bandwidth capacity, makes it a promising complement to previous communication infrastructures such as the Controller Area Network (CAN), the Local Interconnect Network (LIN), or FlexRay. The legacy Ethernet standards were originally targeting non real-time applications and timing-aware features were missing. To fill this gap, several modification standards have been proposed during the last three decades by the Institute of Electrical and Electronics Engineers (IEEE). The collection of all features introduced to this end purpose is refereed to as Time Sensitive Networking (TSN). Among all these, frame preemption, which commands the interruption of a transmitting frame on a specific link of the network for the transmission of another frame considered more urgent on this link, holds a prominent place. The interrupted frame resumes its transmission on the link only after the urgent one has completed. While frame preemption greatly improves the suitability of Ethernet for real-time communication with stringent and heterogeneous requirements, its present operation, as specified by TSN, has some serious limitations that affect network performance. For example, it allows for just one-level preemption. That is, (1) only one class of frames considered as “the most urgent” (usually due to their stringent timing requirements) can preempt all other frames not belonging to this class; and (2) frames belonging to the same class cannot preempt each other. A critical look at this paradigm raises a number of concerns for critical real-time applications. Indeed, there are frames that do not fall into the “most urgent” class, but nevertheless have timing requirements that can only be met if preemption is enabled for frames in the same class and/or the concept of “class” is revisited. By following the current specification, the aforementioned frames can be prevented from transmission for very long time periods due to the transmission of “less urgent” frames and thereby, fail to meet their timing requirements or they may experience very diminished Quality of Service (QoS). Another notable limitation in the standards is the absence of any configuration definition to increase latency performance and limit cases of unmet timing requirements. This work addresses these issues by mitigating the limitations of the one-level preemption scheme, specifically in the IEEE802.1Qbu and IEEE802.3br standards.

Mubarak Ojewale