Overlap Communication in MPI Implementations

Article Sidebar

PDF
Published: Dec 31, 2014
Keywords:
MPI overlap benchmark asynchronous point to point

Main Article Content

Thananon Patinyasakdikul
Innovative Computer Laboratory, University of Tennessee
George Bosilca
Innovative Computer Laboratory, University of Tennessee
Jack Dongarra
1120 Volunteer Blvd University of Tennessee, Knoxville, TN 37916

Abstract

High performance computing has been a very big trend for the past decade. We are in the age that can not just rely on speedup from silicon alone anymore. Computer scientists have been implementing ways to improve the performance of application by making use of the increased of CPU core. One of those technique is “Message Passing Interface (MPI)”, which has a long history back to 1980. MPI standard has been improved since then. Many organizations developed their own implementation of MPI under MPI standard to serve as a tool to achieve higher performance. In this paper, we focus on the basic part of MPI which is point to point communication. We benchmark famous MPI implementations and implement overlap communication for Open MPI. We demonstrate the process, the problem we encountered and the result of our implementation.

Article Details

Issue
Vol. 8 No. 2 (2014): July-December 2014
Section
Academic Article
Creative Commons License

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

References

Ian, P. (2013). The end ofMoore’s law is on the horizon [On-line]. Available:http://www.pcworld.com/article/2032913/the-end-of-moores-law-is-on-the-horizon-says-amd.html

MPI Forum. (2012). MPI: A Message-Passing Interface Standard [On-line]. Available: http://www.mcs.anl.gov/research/projects/mpi/mpi-standard/mpi-report-1.1/mpi-report.html

W. Gropp and E. Lusk and N. Doss and A. Skjellum. Parallel Computing 1996, A high-performance, portable implementation of the {MPI} message passing interface standard. Pittsburgh, Pennsylvania.

Ohio State University. (2002). MVAPICH: MPI over Infini Band, 10GigE/iWARP and RoCE [On-line].

Available: http://mvapich.cse.ohio-state.edu/

Top500.org. (2014). Top 500 Supercomputer Sites [On-line]. Available: http://www.top500.org/

Intel. (2012). Intel MPI Library [On-line]. Available: https://software.intel.com/en-us/intel-mpi-library

Edgar Gabriel, et al. Euro PVM/MPI 2004, Open MPI: Goals, Concept, and design of Next generation MPI Implementation. Budapest, Hungary.