Using Timed Automata for Response Time Analysis of Distributed Real-Time Systems

Abstract

Rate Monotonic Analysis (RMA) is a well-established technique for assessing schedulability of periodic and sporadic tasks which share a processor resource using fixed priority scheduling. Adaptations of this technique have been made to perform Response Time Analysis (RTA), accounting for jitter, blocking, distributed systems and end-to end timing constraints. However, the nature of the analysis means that, while good bounds can be given for uni-processor systems with relatively little interdependency, the response times calculated for more complex systems can be very conservative.An alternative approach to analysing such systems is to build a model which represents the behaviour of the system more dynamically, taking into account the dependency between the tasks. To do this, we introduce a simple language for describing the tasks which comprise a system and the precedence relationships between them. From this a timed hybrid automaton is generated which can be analysed automatically to predict end-to-end response times.Applying this technique in practice yields promising results, with response times lower than those calculated with RTA. However, there is a trade-off to be made between thecomplexity of the hybrid automaton analysis (which suffers from the state explosion problem) and the conservatism of the more standard RTA approach.