Locality-aware PMI usage for efficient MPI startup

Ken Raffenetti, Neelima Bayyapu, Dmitry Durnov, Masamichi Takagi, Pavan Balaji

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

In this paper, we examine usage of the Process Management Interface (PMI) during MPI-Init. Specifically, how PMI is used to exchange address information between peer processes in an MPI job. As node and core counts continue to increase in HPC systems, so does the amount of address data processes need to exchange. We show how by applying well-established locality-awareness techniques, we can significantly reduce the time spent in MPI-Init. We first present the use of shared memory to reduce the total amount of information retrieved from PMI. Next, by combining shared memory with a minimally connected set of processes, we further reduce the dependence on PMI, and employ the HPC fabric to transfer the bulk of address data. Our approach is low impact, relying on functionality already provided by MPI libraries and process managers, instead of new APIs and capabilities.

Original languageEnglish
Title of host publication2018 IEEE 4th International Conference on Computer and Communications, ICCC 2018
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages624-628
Number of pages5
ISBN (Electronic)9781538683392
DOIs
Publication statusPublished - 12-2018
Event4th IEEE International Conference on Computer and Communications, ICCC 2018 - Chengdu, China
Duration: 07-12-201810-12-2018

Publication series

Name2018 IEEE 4th International Conference on Computer and Communications, ICCC 2018

Conference

Conference4th IEEE International Conference on Computer and Communications, ICCC 2018
Country/TerritoryChina
CityChengdu
Period07-12-1810-12-18

All Science Journal Classification (ASJC) codes

  • Computer Networks and Communications
  • Signal Processing

Fingerprint

Dive into the research topics of 'Locality-aware PMI usage for efficient MPI startup'. Together they form a unique fingerprint.

Cite this