Versatile communication algorithms for data analysis

Abstract. Large-scale parallel data analysis, where global information from a variety of problem domains is resolved in a distributed memory space, relies on communication. Three communication algorithms motivated by data analy-sis workloads—merge based reduction, swap based reduction, and neighborhood exchange—are presented, and their performance is benchmarked. These algo-rithms communicate custom data types among blocks assigned to processes in flexible ways, and their performance is optimized by tunable parameters. Perfor-mance is compared with an MPI implementation and with previous communica-tion algorithms on an IBM Blue Gene/P supercomputer at a variety of message sizes and process counts