Allocation of Parallel Programs with Time Variant Resource Requirements

This paper presents an innovative allocation strategy for parallel programs composed of communicating processes whose resource requirements vary during program execution. The primary difference between this and previous work is the incorporation into the allocation process of the estimation of variation in resource requirements over time. Experimental results are presented that compare this model with previous work and indicate that this method achieves superior allocation results without any increase in computational complexity. Results indicate a high correlation between the improvement achieved and the degree of process clustering in the parallel program.