On Task Period Assignment in Multiprocessor Real-Time Control Systems

Abstract

In real-time control systems, a well-known problem is the period assignment to individual tasks, in order to minimize the overall control cost while guaranteeing the task deadlines. In general, the control cost decreases in convex fashion with decreasing periods (increasing invocation rates). Many real-time control systems are increasingly implemented on multiprocessor platforms due to the increased performance requirements. In this paper, we consider the optimal period assignment problem on a homogeneous multiprocessor platform. The problem is intractable in nature. We analyze the performance of the approaches that first partition the tasks, before assigning periods to optimize overall cost on each CPU locally. Then we propose a technique which assigns the periods optimally by reducing the problem to a single-processor problem setting in the first step, and then applying the partitioning algorithms in the second step. Our experimental evaluation shows that the two variants of our proposed technique offer significant advantage, and exhibit a performance close to the theoretical bound achievable by any algorithm.