Efficient processor allocation strategies for mesh-connected parallel computers

Abstract

The processor allocation problem plays a central role in partitionable multiprocessor systems. In this study, we propose two new processor allocation strategies, called the First Fit (FF) and the Best Fit (BF) strategies, for mesh systems. Both strategies can allocate a submesh of the exact size requested for an incoming task, completely avoiding internal fragmentation. Moreover, they can find a free submesh for an incoming task whenever there is one available, eliminating allocation misses. The FF strategy allocates an incoming task to the first available submesh it finds. On the other hand, the BF strategy tries to schedule the task to a submesh in the smallest region of free processors, reserving the larger regions for later tasks. Efficient implementations are presented to keep the searching overhead low. Simulation runs have been conducted to compare the performance of the different allocation schemes.