Date

July 1993

Document Type

Thesis

Degree Name

M.S.

Department

Dept. of Computer Science and Engineering

Institution

Oregon Graduate Institute of Science & Technology

Abstract

There is currently a proliferation of parallel supercomputers made possible by rapidly changing technologies. They have diverse architectures and support different programming models and parallel paradigms. An important and emerging class among parallel architectures are the distributed memory multiprocessors. We focus on the problem of evaluation and the need to provide common performance metrics for these distributed memory machines. Of the existing benchmarks, uniprocessor benchmarks such as Whetstones and Dhrystones are not sufficient for benchmarking distributed memory architectures. Other benchmarks port full-length codes from a range of typical applications. However, since distributed memory machines have a partitioned address space, and communication between the individual processors in the system is implemented via message passing, porting application-level benchmarks to these machines requires extensive restructuring of the original code. It is also difficult to extract fundamental communication characteristics from the information returned by application-level benchmarks. Knowledge of these fundamental characteristics guides users in porting and in making realistic guesses of the performance of their applications on distributed memory systems. In this context, we propose CoMet (Communication Metrics), a synthetic, but comprehensive, benchmark that is easily portable and scalable for distributed memory architectures. CoMet, which is written in C, is based on common communication patterns found in typical parallel scientific algorithms and adds to the information provided by other application-based benchmarks. This thesis overviews existing benchmarks in the context of evaluating distributed memory message-passing machines, presents the CoMet benchmark and describes an implementation of CoMet on Intel's iPSC/860.

Identifier

doi:10.6083/M457190W

Share

COinS
 
 

To view the content in your browser, please download Adobe Reader or, alternately,
you may Download the file to your hard drive.

NOTE: The latest versions of Adobe Reader do not support viewing PDF files within Firefox on Mac OS and if you are using a modern (Intel) Mac, there is no official plugin for viewing PDF files within the browser window.