{"title":"Supercomputing in the South Pacific: performance of a parallel cluster using existing USP facilities","authors":"W. Blanke, Imtiyaz Hussein","doi":"10.1071/SP04016","DOIUrl":null,"url":null,"abstract":"This paper presents the details of a parallel computing cluster built using existing computing resources at the University of the South Pacific. Benchmarking tests using the High Performance Linpack Benchmark were done in order to measure the gigaflops (billions of floating point operations per second) ratings for solving large systems of linear equations while varying the number of computers and Ethernet switches used. These tests provided an overall maximum gigaflops rating which allowed comparison of USP's cluster with leading edge clusters from around the world. Efficiency results also provided insight in how improving the existing network infrastructure might improve the performance of USP's cluster and increase its gigaflops rating. Further tests revealed that the number of Ethernet switches used in USP's current network layout is a definite contributor to the low efficiency of the system as a whole.","PeriodicalId":148381,"journal":{"name":"The South Pacific Journal of Natural and Applied Sciences","volume":"217 1","pages":"0"},"PeriodicalIF":0.0000,"publicationDate":"2004-12-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"The South Pacific Journal of Natural and Applied Sciences","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1071/SP04016","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0
Abstract
This paper presents the details of a parallel computing cluster built using existing computing resources at the University of the South Pacific. Benchmarking tests using the High Performance Linpack Benchmark were done in order to measure the gigaflops (billions of floating point operations per second) ratings for solving large systems of linear equations while varying the number of computers and Ethernet switches used. These tests provided an overall maximum gigaflops rating which allowed comparison of USP's cluster with leading edge clusters from around the world. Efficiency results also provided insight in how improving the existing network infrastructure might improve the performance of USP's cluster and increase its gigaflops rating. Further tests revealed that the number of Ethernet switches used in USP's current network layout is a definite contributor to the low efficiency of the system as a whole.