An adaptive scheduling algorithm for fault-tolerant real-time systems

Abstract

This paper is particularly concerned with the deadline mechanism, proposed as a means to design and implement fault-tolerant real-time systems. The computing system considered here uses a single-processor machine to run multiple software processes; each of them executes a task in response to periodic requests. Based on redundancy, the deadline mechanism assumes that every task is composed of primary and alternate algorithms. In such a context, the scheduling problem that arises amounts to ensuring timely completion of all the requests, either by primaries or alternates but preferably by primaries whenever possible. We present a heuristic scheduling strategy, which is optimal and runs in linear time. Because it is based on the accurate knowledge of processor activity at any point in time, its main property lies in its quick adaptivity and consequently its ability to cope with all runtime situations.