Promotor: prof.dr. J.J. Lukien (TU/e)
Co-promotor: dr.ir. P.J.L. Cuijpers (TU/e)
Eindhoven University of Technology
Date: 6 June, 2023
There is an increasing demand in the automotive industry, high-tech systems, and industrial automation, to accommodate heterogeneous data exchange with high bandwidth communication systems capable of providing real-time guarantees for time-sensitive traffic. The development on top of Ethernet standards enables prioritized data transmission and relies on traffic shaping techniques to allow for different priorities to meet their respective bandwidth and latency needs. We focus on one standardized credit-based shaper that aims to provide the required latency and throughput for audio/video traffic.
Most timing analyses with respect to traffic shaping require a detailed traffic model at each priority level. This dependence on traffic models complicates the analysis, and developers confront the problem of an undesired coupling between different applications, often built by different development teams. Moreover, these analyses are not robust against changes in the system design and are not robust in the event that the system fails to perform according to the specified traffic model.
In this work, we invent an independent timing analysis that considers all traffic from other priorities as inter-priority interference and merely requires shaper-level knowledge rather than a detailed traffic model. Unlike conventional methods, which calculate the worst-case response time of the traffic of interest in a single attempt, we take a relative approach to achieve the same goal, by first computing the worst-case response time in an un-interfered execution and then adding on the relative worst-case response time. This independent analysis not only leads to a preliminary simplification of design concerns, but the resulting estimate can also be proven tight under certain conditions, in contrast to the conventional busy period analyses. We attribute the tightness of our analysis to our introduction of the notion of Eligible Intervals. Unlike busy period analyses, Eligible Intervals are tailored to take the idling nature of credit-based shaping into account.
This dissertation consists of three main parts. Firstly, we present a relative worst-case response time analysis for a credit-based shaper of interest, given the interference consisting of multiple high priorities under credit-based shaping and multiple low priorities. We provide rigorous proof to determine an upper bound where only shaper-level knowledge is relevant, followed by a further investigation of the conditions under which this independent analysis is tight. We also compare our relative worst-case time analysis to the conventional busy period analyses. It turns out that certain sources of pessimism are present in the conventional approaches, which can be remedied by the use of our relative worst-case response time analysis.
Secondly, we use this independent timing analysis to estimate the relative best-case response time. In particular, we show that adding interference prior to a burst of frames in an un-interfered execution contributes to an earlier transmission instead of a delay in the start of frame transmissions.
Thirdly, we use this independent timing analysis as a foundation to determine a minimum bandwidth reservation for credit-based shapers, outperforming previous work based on busy period analyses. Two constraints in determining a minimum bandwidth reservation are derived, namely a deadline constraint and a utilization constraint. In addition, we conduct a set of comparative experiments and demonstrate an improvement in the bandwidth reservation efficiency, i.e., a decrease in the required bandwidth while retaining the independence of the inter-priority interference.