Job-based queue delay modeling in a space-shared hypercube

Processor allocation problem in a space-shared hypercube parallel computer is considered. Since jobs are dynamically allocated and deallocated, existence of more free processors than required does not guarantee the immediate allocation due to the fragmentation problem. If the immediate allocation fails, the processor allocator should decide whether to wait or to assign a smaller subcube to the job. Here, the decision is made depending on the wait or queue delay time. There have been several attempts to predict the queue delay analytically based on the number of busy processors, not the busy subcubes. Main disadvantages of the processor-based prediction is that the number of busy processors does not accurately explain the queue delay. A subcube with the minimum number of running jobs is probably the best candidate for the future allocation even though it has less free processors. We propose a job-based queue delay prediction. We view the allocation moment when all the existing jobs, not the busy processors, in one of the possible subcubes complete their execution. We performed an extensive simulation study and it reveals that the predicted queue delay matches very well with the simulation results. Using the predicted queue delay time, the processor allocator is able to make a right decision and thus, enhances the system performance