Faster topology-aware collective algorithms through non-minimal communication

Paul Sack, William Gropp

Research output: Contribution to journalArticlepeer-review

Abstract

Known algorithms for two important collective communication operations, allgather and reduce-scatter, are minimal-communication algorithms; no process sends or receives more than the minimum amount of data. This, combined with the data-ordering semantics of the operations, limits the flexibility and performance of these algorithms. Our novel non-minimal, topology-aware algorithms deliver far better performance with the addition of a very small amount of redundant communication. We develop novel algorithms for Clos networks and single or multi-ported torus networks. Tests on a 32k-node BlueGene/P result in allgather speedups of up to 6x and reduce-scatter speedups of over 11x compared to the native IBM algorithm. Broadcast, reduce, and allreduce can be composed of allgather or reduce-scatter and other collective operations; our techniques also improve the performance of these algorithms.

Original languageEnglish (US)
Pages (from-to)45-54
Number of pages10
JournalACM SIGPLAN Notices
Volume47
Issue number8
DOIs
StatePublished - Aug 2012

Keywords

  • Collective-communication algorithms

ASJC Scopus subject areas

  • General Computer Science

Fingerprint

Dive into the research topics of 'Faster topology-aware collective algorithms through non-minimal communication'. Together they form a unique fingerprint.

Cite this