M.-S. Wu, R.A. Kendall, and S. Aluru (USA)
Cluster Computing, Collective Communications, MPI, Shared Memory Operations, Tunable Libraries
In this paper we investigate a tunable MPI collective com munications library on a cluster of SMPs. Most tunable collective communications libraries select optimal algo rithms for inter-node communication on a given platform. We add another layer of intra-node communications com posed by several tunable shared memory operations. We explore the advantages of our approach, and discuss when to use our approach, when to switch to another approach on the shared memory layer. Experimental results indi cate that collective communications designed by such an approach with proper tuning can outperform vendor imple mentations.
Important Links:
Go Back