Optimal server allocation for real time computing systems with bursty priority jobs

Abstract

Among the many distinguishing features of the real-time computing system, the response time might be the most important one. In this paper we consider a MIMD type parallel computer system with different job priority and bursty job arrival pattern. Real-time jobs have higher priority than other jobs to guarantee faster response time, and the job server consists of array of processors, i.e. processor pool. Job requests arrive at different request buffers according to their priority, and we assume that the arrival pattern is a bursty process. These two issues, job priority and the bursty job request pattern, are studied for real-time computing systems. The processor pool is shared in such a way that if high priority queue is not empty, a fixed amount of the processing capacity is allocated to the queue, while the rest of the server is allocated to the other queue. Average waiting time is calculated for both queues. The optimal partition ratio of the processor pool is obtained analytically based on the average waiting time of high priority queue and low priority queue with respect to variable values of traffic loads /spl rho/.