1979Unpublished venueOpen access

Evaluating computer program performance on the CRAY-1

L.E. Rudsinski, Gail W. Pieper

Open full text 3 citations

Abstract

The Advanced Scientific Computers Project of Argonne's Applied Mathematics Division has two objectives: to evaluate supercomputers and to determine their effect on Argonne's computing workload. Initial efforts have focused on the CRAY-1, which is the only advanced computer currently available. Users from seven Argonne divisions executed test programs on the CRAY and made performance comparisons with the IBM 370/195 at Argonne. This report describes these experiences and discusses various techniques for improving run times on the CRAY. Direct translations of code from scalar to vector processor reduced running times as much as two-fold, and this reduction will become more pronounced as the CRAY compiler is developed. Further improvement (two- to ten-fold) was realized by making minor code changes to facilitate compiler recognition of the parallel and vector structure within the programs. Finally, extensive rewriting of the FORTRAN code structure reduced execution times dramatically, in three cases by a factor of more than 20; and even greater reduction should be possible by changing algorithms within a production code. It is condluded that the CRAY-1 would be of great benefit to Argonne researchers. Existing codes could be modified with relative ease to run significantly faster than on the 370/195. More important, the CRAY would permit scientists to investigate complex problems currently deemed infeasibile on traditional scalar machines. Finally, an interface between the CRAY-1 and IBM computers such as the 370/195, scheduled by Cray Research for the first quarter of 1979, would considerably facilitate the task of integrating the CRAY into Argonne's Central Computing Facility. 13 tables.

Open-access reader

About this research paper

What this paper is about

The Advanced Scientific Computers Project of Argonne's Applied Mathematics Division has two objectives: to evaluate supercomputers and to determine their effect on Argonne's computing workload. Initial efforts have focused on the CRAY-1, which is the only advanced computer currently available. Users from seven Argonne divisions executed test programs on the CRAY and made performance comparisons with the IBM 370/195 at Argonne. This report describes these experiences and discusses various techniques for improving run times on the CRAY. Direct translations of code from scalar to vector processor reduced running times as much as two-fold, and this reduction will become more pronounced as the CRAY compiler is developed. Further improvement (two- to ten-fold) was realized by making minor code changes to facilitate compiler recognition of the parallel and vector structure within the programs. Finally, extensive rewriting of the FORTRAN code structure reduced execution times dramatically, in three cases by a factor of more than 20; and even greater reduction should be possible by changing algorithms within a production code. It is condluded that the CRAY-1 would be of great benefit to Argonne researchers. Existing codes could be modified with relative ease to run significantly faster than on the 370/195. More important, the CRAY would permit scientists to investigate complex problems currently deemed infeasibile on traditional scalar machines. Finally, an interface between the CRAY-1 and IBM computers such as the 370/195, scheduled by Cray Research for the first quarter of 1979, would considerably facilitate the task of integrating the CRAY into Argonne's Central Computing Facility. 13 tables.

Why it matters

OpenAlex reports 3 citations for this work. Citation counts describe recorded attention and do not establish research quality.

Key contribution

A contribution statement is not available in the OpenAlex record.

Method / approach

Method details are not available in the OpenAlex metadata.

Main findings

Findings are not separately available in the OpenAlex metadata.

Limitations

Limitations are not available in the OpenAlex metadata.

Applications

Application details are not available in the OpenAlex metadata.

Available abstract

The Advanced Scientific Computers Project of Argonne's Applied Mathematics Division has two objectives: to evaluate supercomputers and to determine their effect on Argonne's computing workload. Initial efforts have focused on the CRAY-1, which is the only advanced computer currently available. Users from seven Argonne divisions executed test programs on the CRAY and made performance comparisons with the IBM 370/195 at Argonne. This report describes these experiences and discusses various techniques for improving run times on the CRAY. Direct translations of code from scalar to vector processor reduced running times as much as two-fold, and this reduction will become more pronounced as the CRAY compiler is developed. Further improvement (two- to ten-fold) was realized by making minor code changes to facilitate compiler recognition of the parallel and vector structure within the programs. Finally, extensive rewriting of the FORTRAN code structure reduced execution times dramatically, in three cases by a factor of more than 20; and even greater reduction should be possible by changing algorithms within a production code. It is condluded that the CRAY-1 would be of great benefit to Argonne researchers. Existing codes could be modified with relative ease to run significantly faster than on the 370/195. More important, the CRAY would permit scientists to investigate complex problems currently deemed infeasibile on traditional scalar machines. Finally, an interface between the CRAY-1 and IBM computers such as the 370/195, scheduled by Cray Research for the first quarter of 1979, would considerably facilitate the task of integrating the CRAY into Argonne's Central Computing Facility. 13 tables.

Key concepts: Computer science, Compiler, Supercomputer, Fortran, Parallel computing, National laboratory, IBM, Operating system

Related papers

Back to paper searchBrowse research topicsOriginal source
Evaluating computer program performance on the CRAY-1 — Research Paper | ScholarLens