Improving processor availability in the MPI implementation for the ASCI/Red supercomputer

Author

Brightwell, Ron ; Lawry, William ; Maccabe, Arthur B. ; Wilson, Christopher

Author_Institution

Scalable Comput. Syst. Dept., Sandia Nat. Labs., Albuquerque, NM, USA

fYear

2002

fDate

6-8 Nov. 2002

Firstpage

639

Lastpage

647

Abstract

This paper describes how a portable benchmark suite that measures the ability of an MPI implementation to overlap computation and communication can be used to discover and diagnose performance problems. We describe the approach of the benchmark suite and discuss a performance problem that we uncovered with the MPI implementation on the ASCI/Red supercomputer. A slight modification to the MPI implementation has resulted in a significant gain in CPU availability and bandwidth with a slight degradation in latency performance. We present a detailed analysis of these results and discuss how the benchmark suite has enabled us to tailor the MPI implementation to optimize for all three measurements.

Keywords

delays; message passing; parallel machines; performance evaluation; ASCI/Red supercomputer; CPU availability; MPI implementation; bandwidth; latency performance; performance diagnosis; portable benchmark suite; processor availability; Availability; Bandwidth; Computer networks; Computer science; Contracts; Degradation; Laboratories; Performance gain; Portable computers; Supercomputers;

fLanguage

English

Publisher

ieee

Conference_Titel

Local Computer Networks, 2002. Proceedings. LCN 2002. 27th Annual IEEE Conference on

ISSN

0742-1303

Print_ISBN

0-7695-1591-6

Type

conf

DOI

10.1109/LCN.2002.1181841

Filename

1181841