Early-stage Bottleneck Identification and Removal in TSN Networks
Prof. Nicolas Navet , University of Luxembourg;
Dr. Hoai Hoang Bengtsson , Volvo;
Dr. Jörn Migge , REALTIME-AT-WORK (RTaW)
A main issue in the design of automotive networks is that important design choices pertaining to the topology and the technologies (i.e., protocols, data rate, hardware) have to be made at a time when the communication needs are not entirely known. The problem becomes more important for next-generation cars, as in-vehicle communication requirements will evolve and grow with new software functions and services being added during the lifetime of the vehicle. It is thus crucial to identify and remove bottlenecks in a candidate communication architecture so as to make the best possible use of the hardware resources and the technologies chosen. This will allow OEMs to offer the best possible service to the customers, generate revenues with new functions and services, and extend the lifetime of the platform. In this study we tackle the problem in the context of TSN backbone networks supporting mixed criticality streams (legacy traffic: audio, video, control, adas, fusion, data) and Some/IP services with different types of performance contraints (throughput and hard/soft deadlines).
In the following, a TSN configuration refers to a TSN network which has been fully configured: network layout, data rates, streams allocated to the traffic classes, scheduling mechanism chosen for all traffic classes, etc. Our approach consists of three steps which can be automated by algorithms:
1. Quantify the "capacity" of a candidate TSN configuration in terms of the number of streams and services of the different types it can schedule,
2. Identify bottleneck resources in the TSN network, which can be links (too low date rates) and switches (switching delay, available memory, sub-optimal TAS schedule, etc),
3. Propose incremental improvements (e.g. duplicate a link, increase memory) and rank them in terms of how much they improve the capacity of the network.
In our experience on a realistic centralized SOA architecture a single underdimensioned 100Mbit/s link can diminish the capacity of the TSN configuration in terms of number of services that the E/E architecture can host by 40%!
We introduce the concepts of "bottleneck constraints", the most limiting constraints (e.g. a 10Mbit/s throughput constraint for software update) for the network and show to identify them. We then discuss what a bottleneck resource is, and show that the load of a resource is insufficient to identify bottlenecks. We propose new metrics based the contribution of a resource to the violation of the performance constraints.
Throughout the presentation, the key concepts and steps of the approach are examplified on a realistic next-generation high-performance centralized architecture. This architecture is used to illustrate how the approach advocated allows to improve early-stage design choices and contribute to the design of more future-proof TSN networks. The aim of this presentation is also to raise discussions among the participants about the future of tool-assisted design.