Method for Maximal Utilization of Idle Links for Fast Load Balancing
Hwa-Gyeong Im, Ju-Uk Jang, Seong-Cheon Kim · Jeongbo gwahaghoe nonmunji. si'seu'tem mich i'lon · 2001
In this paper, we introduce new methods for hiding computation overheads involved in load redistributing for parallel computer of hypercube, mesh and tree topologies. The basic idea is either coalescing some phases of load redistributing to overlap the transfer on different links or dividing each phase into steps to pipeline the transfer of load unit by unit for maximum utilization of links. They proved effective in making links busy transmitting load as soon as possible, hence reducing the computation overheads involved in balancing. Proposed techniques experimented on hypercube, mesh or tree topologies reduce communication overheads by 20% to 50% compared with known methods.