An empirical study of optimization in seamless remote MPI-I/O for long latency network

Authors:
Yuichi Tsujita
Affiliations:
Department of Electronic Engineering and Computer Science, School of Engineering, Kinki University, Higashi-Hiroshima, Hiroshima, Japan
Venue:
PVM/MPI'07 Proceedings of the 14th European conference on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Year:
2007

Citing 4
Cited 0

A high-performance, portable implementation of the MPI message passing interface standard

Parallel Computing
On implementing MPI-IO portably and with high performance

Proceedings of the sixth workshop on I/O in parallel and distributed systems
An Architecture of Stampi: MPI Library on a Cluster of Parallel Computers

Proceedings of the 7th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Stampi-I/O: A Flexible Parallel-I/O Library for Heterogeneous Computing Environment

Proceedings of the 9th European PVM/MPI Users' Group Meeting on Recent Advances in Parallel Virtual Machine and Message Passing Interface

Quantified Score

Hi-index	0.00

Visualization

Abstract

A Stampi library realizes MPI-I/O operations among computers which have different MPI libraries by bridging the both libraries with TCP sockets. If interconnections among computers have long latency, throughput is degraded due to unoptimized configuration of TCP sockets. For effective remote MPI-I/O operations, improvement in throughput is an important issue. In this research work, I/O performance was measured on interconnected PC clusters which were in different network segments to find desirable configuration. We notice that optimization in socket buffer sizes and precise traffic control were quite effective to achieve high throughput I/O.