cheng@pollux.usc.edu (Chien-Ming Cheng) (11/09/90)
Does anyone know whether there is any benchmark experiments for the hypercube machine? Pointers to research papers are also welcome. Ted Cheng
picano@en.ecn.purdue.edu (Silvio Picano) (11/10/90)
In article <28021@usc> cheng@pollux.usc.edu (Chien-Ming Cheng) writes: > >Does anyone know whether there is any benchmark experiments for >the hypercube machine? >Pointers to research papers are also welcome. > > >Ted Cheng D. C. Grunwald, D. A. Reed, \fIBenchmarking Hypercube Hardware and Software\fR, No. UIUCDCS-R-86-1303, Urbana, IL, Nov. 1986, pp. 1-17. J. Gustafson, G. Montry, R. Benner, \fIDevelopment of Parallel Methods for a 1024-Processor Hypercube\fR, SIAM, Vol. 9, No. 4, July 1988. From pur-ee!mentor.cc.purdue.edu!purdue!tut.cis.ohio-state.edu!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!emory!hubcap!kulcs!dirkr Wed Apr 4 16:24:36 EST 1990 Article 1314 of comp.parallel: Path: en.ecn.purdue.edu!pur-ee!mentor.cc.purdue.edu!purdue!tut.cis.ohio-state.edu!zaphod.mps.ohio-state.edu!sol.ctr.columbia.edu!emory!hubcap!kulcs!dirkr >From: kulcs!dirkr@relay.EU.net (Dirk Roose) Newsgroups: comp.parallel Subject: benchmark programs for iPSC/2 Keywords: benchmarks hypercube iPSC Message-ID: <8622@hubcap.clemson.edu> Date: 4 Apr 90 14:45:52 GMT Sender: fpst@hubcap.clemson.edu Lines: 35 Approved: parallel@hubcap.clemson.edu We developed some programs for benchmarking the iPSC/2, including benchmarks on - integer calculation and floating point calculation - dhrystone and whetstone benchmarks - nearest neigbour communication - non-neigbour communication (multi-hop communication) - bidirectional communication (exchange data) - channel contention measurement (with and without "hot spot") - SRM - node0 communication - Remote Host - node0 communication The results of these programs (for NX/2 Version 2.2 and 2.4) are presented in: - L.Bomans and D. Roose, Benchmarking the iPSC/2 hypercube multiprocessor, Concurrency: Practice and Experience, 1, pp. 93-103, 1989. - L.Bomans and D. Roose, Communication Benchmarks for the iPSC/2, in Hypercube and Distributed Computers, F. Andre and J.P. Verjus (ed.), North-Holland, 1989, p. 93-103. The programs are all written in C (except one Fortran program). Shell-scripts are provided to run the benchmarks and to collect the results in a "report". If you are interested in obtaining these benchmark programs, send me an e-mail message. Then you will receive the programs (as an "uucoded" Unix shell archive), together with the results for NX/2 version 3.0. Extension of the code and modification for use on other distributed memory machines is encouraged! Dirk Roose Dept. of Computer Science E-MAIL : dirkr@cs.kuleuven.ac.be K. U. Leuven dirkr@kulcs.BITNET Celestijnenlaan 200A dirkr@kulcs.UUCP B-3030 Leuven PHONE : +32 16 20 06 56 x3546 BELGIUM FAX : +32 16 20 53 08 From pur-ee!mentor.cc.purdue.edu!noose.ecn.purdue.edu!samsung!cs.utexas.edu!usc!brutus.cs.uiuc.edu!dali!ogicse!emory!hubcap!dbradley Wed Apr 4 16:26:57 EST 1990 Article 1317 of comp.parallel: Path: en.ecn.purdue.edu!pur-ee!mentor.cc.purdue.edu!noose.ecn.purdue.edu!samsung!cs.utexas.edu!usc!brutus.cs.uiuc.edu!dali!ogicse!emory!hubcap!dbradley >From: dbradley@gibson.ncsa.uiuc.edu (David Bradley) Newsgroups: comp.parallel Subject: Re: benchmark programs for iPSC/2 Keywords: benchmarks hypercube iPSC Message-ID: <8627@hubcap.clemson.edu> Date: 4 Apr 90 19:03:22 GMT Sender: fpst@hubcap.clemson.edu Lines: 28 Approved: parallel@hubcap.clemson.edu In article <8622@hubcap.clemson.edu> kulcs!dirkr@relay.EU.net (Dirk Roose) writes: >We developed some programs for benchmarking the iPSC/2, including benchmarks on > - integer calculation and floating point calculation > - dhrystone and whetstone benchmarks > - nearest neigbour communication > - bidirectional communication (exchange data) These benchmarks are also included in my MS Thesis "First and Second Generation Hypercube Performance", available from the University of Illinois Computer Science Department as Tech Report UIUCDCS-R-88-1455, September 1988. The tests were run under NX 2.2; version 2.4 was not available back then. In addition to C processor benchmarks, FORTRAN performance is also anyalyzed, including performance of the Livermore Loops, and FORTRAN Whetstones. Several programs were coded in both C and FORTRAN, and the relative performance of the two compilers was compared. On the communication front, in addition to the simple benchmarks mentioned above, the effects of spatial and temporal locality in message traffic was examined, as were as the effect of hot spots. Finally, the performance of distributed LINCUBE and Simplex algorithms is included. The report includes measurements for the iPSC/1 and Ncube, as well as for the iPSC/2. -- David Bradley University of Illinois at Urbana Champaign Silvio