EFFECT OF QUEUE SELECTION AND SERVICE TIME DISTRIBUTIONS IN MULTIPLE RESOURCE-ALLOCATION

This paper examines a system with a number of tasks and a set of identical (commodity-type) resources. The resources may be, for example, processors, buffers or communication channels. At some instant in time, each task requires a number of resources before it can resume execution. The task request is entered in a queue, the index of which indicates the number of requested resources. If the resources are available, they are allocated and the task is served immediately; otherwise the request remains in the queue. When more than one request can be served due to resource availability, the one with the highest priority is chosen first. After service, each task stays in a processing stale for a period of time before it generates the next request. This paper develops a Markov-chain based solution to a model of such systems, and calculates the system state probabilities and performance measures. Queue selection probabilities and expected service times are assumed to depend on the queue index. The behaviour of these systems is examined when higher priority is assigned to requests in the smaller indexed or the larger indexed queues.