CoMPI: configuration of collective operations in LAM/MPI using the scheme programming language

Authors:
Espen Skjelnes Johnsen;John Markus Bjørndalen;Otto J. Anshus
Affiliations:
Dept. of Computer Science, University of Tromsø, Tromsø, Norway;Dept. of Computer Science, University of Tromsø, Tromsø, Norway;Dept. of Computer Science, University of Tromsø, Tromsø, Norway
Venue:
PARA'06 Proceedings of the 8th international conference on Applied parallel computing: state of the art in scientific computing
Year:
2006

Citing 3
Cited 1

Revised5 report on the algorithmic language scheme

ACM SIGPLAN Notices
MagPIe: MPI's collective communication operations for clustered wide area systems

Proceedings of the seventh ACM SIGPLAN symposium on Principles and practice of parallel programming
CC--MPI: a compiled communication capable MPI prototype for ethernet switched clusters

Proceedings of the ninth ACM SIGPLAN symposium on Principles and practice of parallel programming

Modeling resource-coupled computations

Proceedings of the 2009 Workshop on Ultrascale Visualization

Quantified Score

Hi-index	0.01

Visualization

Abstract

This paper describes CoMPI, an extension to LAM/MPI, which enables the user to configure collective operations by using the high level programming language Scheme. A high level language is used to achieve flexibility. When using CoMPI, applications do not have to be modified or recompiled. We implemented the allreduce collective operation in Scheme, preserving the original LAM/MPI algorithm as implemented in C. We measured the performance and found that the overhead using Scheme was insignificant. Using CoMPI to reconfigure the allreduce communication pattern taking the network topology into account, significant improvement in performance on a multi-cluster was achieved.