Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!husc6!uwvax!oddjob!ncar!ames!amdahl!chuck From: chuck@amdahl.uts.amdahl.com (Charles Simmons) Newsgroups: comp.arch Subject: Re: CISC strikes back. Message-ID: <31755@amdahl.uts.amdahl.com> Date: 13 May 88 18:34:15 GMT References: <30872@amdahl.uts.amdahl.com> <3460014@hpsrla.HP.COM> <2175@winchester.mips.COM> Reply-To: chuck@amdahl.uts.amdahl.com (Charles Simmons) Organization: Amdahl Corporation, Sunnyvale CA Lines: 30 In article <2175@winchester.mips.COM> mash@winchester.UUCP (John Mashey) writes: >This is one more reminder that we really could use some more good, realistic, >substantive, non-toy, public domain, commonly-cited integer benchmarks. >It would be awfully nice to get at least 6-10 good ones that people >agreed had some representative qualities. >-- >-john mashey DISCLAIMER: >UUCP: {ames,decwrl,prls,pyramid}!mips!mash OR mash@mips.com >DDD: 408-991-0253 or 408-720-1700, x253 >USPS: MIPS Computer Systems, 930 E. Arques, Sunnyvale, CA 94086 Possibly this is an inappropriate newsgroup for discussing benchmarks. If so, I'm sure all you kind people will send me mail so that I'll never talk about benchmarks here again. Since John has quite a bit of experience working with benchmarks, I'd be interested in hearing his thoughts on what makes a benchmark good, realistic, and substantive. Maybe other people would like to contribute their ideas as well? As a question, can a benchmark be good and useful if it runs in time proportional to N*N where N is an input parameter? For example, the drhystone benchmark runs in time c*N where N is the number of loops to execute. This makes it very easy to compare dhrystone results obtained from machines where a different number of loops were run on each different machine. For a more complex program like 'sieve', it might be difficult to obtain comparable results across a broad class of machines. Thanks, Chuck