Performance evaluation of parallel processing systems using queueing network model

This paper introduces a queueing model to measure the performance of parallel processing network. The prototype system with 80 parallel computers for all subtasks execution is implemented. The case in which a task with granularity is discussed. The parallel system apparently more or less will improve the performance by distributing and executing subtasks on dedicated 80 parallel computers. Delays in task partitioning, subtask distribution, communications and merging are taken into account for this performance evaluation. However, the prototype of this parallel environment is indeed tedious and costly. In this paper an alternative, cost-effective queueing model is presented in order to investigate how parallel systems can function, as tasks with granularity exist. Results from cost-effective model are compared vis-à-vis results captured by prototype system with 80 processors. By varying task characteristics for the mentioned prototype system, the supportive results from the cost-effective queueing model can be derived. In addition, a performance evaluation for a model with 200 processors can be predicted by this proposed method.