Predicting execution time of real-time programs on contemporary machines

A computer program that interacts with and reponds to real world processes in a timely fashion, and must com­ plete execution prior to its scheduled deadline, is called a ’’hard” real-time program. It is not sufficient for the implemented algorithm to be correct. The real-time pro­ gram must provide the correct response (computation) on time. A late computation is usually no better, possi­ bly even worse, than one that is on time but imprecise. The timing behavior of each real-time program compo­ nent (task) must be predictable if one is to build reliable deterministic real-time systems. Much of the research in real-time scheduling assumes that the execution lime of each task is known (e.g., Liu and Layland [2]). Work by Mok [3], on the use of semaphores, rendezvous, and monitors in real-time systems assumes that the (worst case) execution time of code segments is known. Stoyenko’s work [8] on the schedulability analyzer for Real-Time Euclid addressed the problem of worst case timing analysis of tasks, by assuming the execution time of each instruction is con­ stant. However, even the hardware builders [4] concede that the exact execution time of a given instruction may vary, depending upon the surrounding instructions and the current 6tate of the machine. Park and Shaw [5] implemented a timing tool for a subset of C, which is based on the notion of a timing schema presented in [7]. A method very similar to that of Shaw is presented by Puschner and Koza [6]. Shaw acknowledges that al­ though his approach seems to work well when applied to determinisitic hardware, more research is needed to determine timing predictability on contemporary ma­ chines.