A Shared Memory Dispatching Approach for Partially Clairvoyant Schedulers

Abstract

It is well known that in a typical real-time system, certain parameters, such as the execution time of a job, are not fixed numbers. In such systems, it is common to characterize the execution time as a range-bound interval, say, [l,u], with l indicating the lower bound on the execution time and u indicating the upper bound on the same. Such intervals can be determined with a high degree of confidence in state of the art operating systems, such as MARUTI [7,5] and MARS [2]. Secondly, jobs within a real-time system are often constrained by complex timing relationships. In hard real-time applications, it is vital that all such constraints are satisfied at run time, regardless of the values assumed by environment-dependent parameters, such as job execution times. As described in [11], there are two fundamental issues associated with real-time scheduling, viz., the schedulability query and dispatchability. A positive answer to the schedulability query may not by itself guarantee that all the imposed constraints will be met at run-time; indeed the phenomenon in which the dispatcher fails to dispatch a schedulable job set, is called Loss of Dispatchability. This paper is concerned with techniques to address this phenomenon in Partially Clairvoyant schedulers; we primarily focus on distributing the dispatch computations across the processors of a shared-memory computer.