HomePublicationsPhD Thesis

On Preemption in Time-Sensitive Networking (TSN)
Ref: CISTER-TR-230202       Publication Date: 28, Feb, 2023

Abstract:
A key requirement for responsiveness, i.e., timely and correct response to events, in real-time distributed applications is the ability of data to move in a reliable and predictable manner across the underlying communication network. Ethernet is the emerging communication technology for modern real-time wired networks. Its ability to meet the increasingly stringent speed and distance requirements, along with its high bandwidth capacity, makes it a promising addition to legacy communication infrastructures. The older Ethernet standards were originally geared towards non-real-time applications and lacked features to support real-time communication. To close this gap, the Institute of Electrical and Electronics Engineers (IEEE) has proposed several changes to standards over the past three decades. The collection of recent standards introduced for this purpose is called Time Sensitive Networking (TSN). Among all these, the IEEE 802.1Qbu, which specifies a 1-level frame preemption mechanism, occupies a prominent place. Specifically, the TSN frame preemption mechanism is specified by the so-called 1-level preemption scheme as follows. Frames are divided into two classes: (i) the express frames, which are considered urgent and therefore eligible for expedited transmission, and (ii) the preemptable frames, which are considered less urgent. In particular, express frames can preempt preemptable frames and two frames of the same class cannot preempt each other.
While frame preemption significantly improves the suitability of Ethernet for real-time communication with strict and heterogeneous requirements, the current mode of operation, as specified by TSN, has some serious limitations that affect network performance. Most importantly, this scheme is prone to performance degradation when the number of express frames is high. In addition, preemptable frames with firm timing requirements can suffer from long blocking periods due to priority inversions, since frames in the same preemption class cannot preempt each other. These limitations mean that the 1-level preemption scheme does not provide a way to efficiently support the coexistence of flows with diverse timing requirements on the same network. In this thesis, we show the limitations of 1-level preemption in real-time applications. We then postulate that these shortcomings can be effectively mitigated with a multi-level preemption scheme, and we show the feasibility of this scheme as well as its requirements, timing analysis, and configuration. We make a fourfold contribution: (1) We propose a new framework in which the non-preemptive transmission constraints between non-express frames are relaxed. Then, we describe the operational dynamics of our approach and the actual implementation recommendations for its feasibility; (2) we perform a comprehensive and rigorous worst-case traversal time (WCTT) analysis for each flow in the network and compare the results with the 1-level preemption and the non-preemptive schemes; (3) we provide an offline priority assignment scheme for the flow set. Then, we provide an offline framework for determining the appropriate number of preemption levels on the one hand and assigning flows to preemption classes on the other; finally (4) we evaluate the performance gain of a multi-level priority scheme over a 1-level preemption scheme from both qualitative and quantitative perspectives. Several studies have pointed out that the major limitation of frame preemption is that it only allows one level of preemption. By addressing this limitation, this work positions Ethernet TSN with Frame Preemption as a simpler and more cost-effective alternative communication solution for Distributed Real-time Embedded Systems.

Mubarak Ojewale

Document:

View
PDF Download
PDF