Multiprocessor schedulability analyser

Abstract

Translate Article

Within the context of hard real-time systems, the schedulability analysis of a task set is a major issue. The problem consists in proving that the tasks always satisfy their temporal constraints for a given scheduling policy and a given platform. Extensive work has been done in the last decades for defining sufficient criteria and exact algorithms. Sufficient criteria usually have an excellent complexity but often lead to an over-dimension of the system. On the opposite, exact algorithms, especially in the case of multiprocessor platform, suffer from an exponential complexity. In this paper, we study an exact technique: we apply a brute force search combined with a model checker (Uppaal) that determines whether the exploration is complete. We consider periodic tasks which execute on parallel platforms composed of homogeneous processors. Under these hypotheses, we have encoded four policies: fixed task priority, gEDF, gLLF and LLREF. The analyser is user friendly and provides promising performances.