Approximate analysis of distributed semi-hard real-time systems

Abstract

In this note we investigate systems which combine the timing requirements of real-time processing with the complexity of knowledge processing. In such real-time environments it cannot be guaranteed that all processes meet their deadlines but it is possible to isolate a subset of critical processes which must meet their deadlines. We call such systems semi-hard real-time control systems. Although priority scheduling is used, the execution time of critical processes has to be determined by a nondeterministic analysis which takes into account contention for system resources with noncritical processes. We investigate two possible architectures for a distributed system in which each process runs on its own processor and they communicate either by messages or through a shared memory. In the latter case a subset of system resources does not allow preemption, for example, a process running into a critical section must complete its function. A methodology for system modeling and the approximations needed are discussed. Even in the simple case, the analysis is nontrivial. Due to priority scheduling the queueing models are nonseparable, hence, product form solutions do not exist.