Implementing reliable distributed real-time systems with the Θ-model

Abstract

A widely accepted viewpoint is that designs for distributed real-time systems should be based on synchronous computational models. Safety in such designs, however, requires that the target system behaves as the synchronous model postulates. We believe that this approach is rather risky, as it rests on solving distributed scheduling problems which are known to be NP-hard. We therefore advocate the use of more relaxed system models, namely asynchronous models equipped with unreliable failure detectors. To this end, we introduce a novel implementation of the perfect failure detector, resting on an abstract model without upper bounds on end-toend message delays. Then, we demonstrate how this algorithm can be transferred from the abstract model into a real network/system architecture. Finally, we prove that this solution exhibits real-time behavior. © Springer-Verlag Berlin Heidelberg 2006.