Performance analysis of dynamic multitasking imprecise computation system

Abstract

The scheduling of tasks in dynamic multitasking computer systems using imprecise partial computations is studied. In this system, tasks arrive randomly during run-time, and they are to be processed as quickly as possible. The task is modelled in such a way that each task has two parts of computation: a mandatory part, and an optional part. The tasks are scheduled such that: if the total number of tasks in the system is no more than M (a system parameter), the tasks are executed fully to produce precise results. When the total number of tasks exceeds M, then only the mandatory parts of the tasks are executed. To study this dynamic imprecise computation system, the authors propose three performance metrics to measure responsiveness and quality of computation of tasks. Queueing methods are used to analyse the scheduling problem, and performance metrics are computed numerically. The performance of the scheme is assessed, and found to be effective in keeping the mean task waiting time at a low value when the system is heavily loaded or overloaded.