PERFORMANCE ANALYSIS OF TASK SYSTEMS USING A QUEUEING NETWORK MODEL.

Abstract

Queuing network (QN) solvers, and especially packages developed for the performance analysis of multiprogrammed computer systems, can only handle independent jobs (in possibly different categories). These packages are not applicable to the analysis of workloads that have precedence relationships among the set of jobs to be processed. Such workloads can be represented by (acyclic) task systems, where each task is specified by its service demands on the target computing system. The authors propose a solution method based on hierarchical decomposition to efficiently compute the performance measures of interest in task systems, such as its mean completion time. An example of the application of the solution method to a load balancing problem is presented.