Contention-free 2D-mesh cluster allocation in hypercubes

Traditionally, each job in a hypercube multiprocessor is allocated with a subcube so that communication interference among jobs may be avoided. Although the hypercube is a powerful processor topology, the 2D mesh is a more popular application topology. This paper presents a 2D-mesh cluster allocation strategy for hypercubes. The proposed auxiliary free list processor allocation strategy can efficiently allocate 2D-mesh dusters without size constraints, can reduce average job turnaround time compared with that based on subcube allocation strategies, and can guarantee no communication interference among allocated clusters when the underlying hypercube implements deadlock-free E-cube routing. The proposed auxiliary free list strategy can be easily implemented on hypercube multicomputers to increase processor utilization.<
>