The effect of real workloads and stochastic workloads on the performance of allocation and scheduling algorithms in 2D mesh multicomputers

The performance of the existing non-contiguous processor allocation strategies has been traditionally carried out by means of simulation based on a stochastic workload model to generate a stream of incoming jobs. To validate the performance of the existing algorithms, there has been a need to evaluate the algorithms' performance based on a real workload trace. In this paper, we evaluate the performance of several well-known processor allocation and job scheduling strategies based on a real workload trace and compare the results against those obtained from using a stochastic workload. Our results reveal that the conclusions reached on the relative performance merits of the allocation strategies when a real workload trace is used are in general compatible with those obtained when a stochastic workload is used.