Path: utzoo!utgpu!jarvis.csri.toronto.edu!cs.utexas.edu!sun-barr!newstop!sun!khb From: khb@chiba.Eng.Sun.COM (Keith Bierman - SPD Advanced Languages) Newsgroups: comp.unix.aix Subject: Re: Risc System/6000 Message-ID: Date: 22 Feb 90 02:12:43 GMT References: <1514@fcs280s.ncifcrf.gov> <4115@ibmpa.UUCP> <10307@hoptoad.uucp> <1990Feb21.012432.22401@ico.isc.com> <415@peyote.cactus.org> Sender: news@sun.Eng.Sun.COM Organization: Sun MegaSystems Lines: 35 In-reply-to: woan@peyote.cactus.org's message of 21 Feb 90 06:43:55 GMT In article <415@peyote.cactus.org> woan@peyote.cactus.org (Ronald S. Woan) writes: you get a system at $650/SPEC or so. Why are SPECs the geometric mean rather than average by the way? Because the arithmetic mean (the "average") isn't reasonable when comparing different rates. I've done a writeup or two, but I've misplaced my brain. Let it suffice to note that: time A .. vs time (A+..+N) time N and compute the average(s). With the arithmetic mean, the single number is jerked up a lot by one big outlier. The harmonic mean produces the same value, which is argued fair by the original livermore loops report, and other folks ... but which turns out to penalize a system for screwing up on one particular code by a lot. The geometric mean falls somewhere between. When I do the algebra, I pick the harmonic mean. When I stare at enough benchmark figures, the geometric mean seems more sensible. Others get different results. I wasn't there for the SPEC meetings, but I bet this was debated for a bit and decided along Mashey's lines. -- Keith H. Bierman |*My thoughts are my own. !! kbierman@Eng.Sun.COM It's Not My Fault | MTS --Only my work belongs to Sun* kbierman%eng@sun.com I Voted for Bill & | Advanced Languages/Floating Point Group Opus | "When the going gets Weird .. the Weird turn PRO" "There is NO defense against the attack of the KILLER MICROS!" Eugene Brooks