eugene@eos.arc.nasa.gov (Eugene Miya) (12/21/90)
In article <115440004@hpcuhc.cup.hp.com> spuhler@hpcuhc.cup.hp.com (Tom Spuhler) writes: ># >for your faster CPU's? Does management want a richer instruction mix to ># ># Er, sorry, I must be dense, but where does the "richer instruction mix"(tm) ># come in (sounds like coffee, thank god I drink tea). Seems like more of > >Come on, Eugene, you're tripping over the easy ones:-) Richer >instruction mix means a more varied, or using a larger subset of the machine >insructions. Well, sorry, Tom. I just didn't see this as improved. You didn't convince me. See I am a skeptic (and cynic). I just see it as more of the same. Business as usual, in an area which needs a new business. The way to win is not to play the game.... >Not particullary interesting, as the important criteria is >how the tested instruction mix matches your expected workloads(for richer SCORE ANOTHER POINT FOR REPRESENTATIVENSS. >Optimization. How likely is someone likely to do something like >that? Depends on how hung up the world gets on a single benchmark. >How likely is someone going to optimize for Dhrystone? (Seems to have >hppened). It's all a matter of contest. .. >We always have to live with imperfect information. You haven't been talking to a particular physicist at the Supercomputer Research Center (SRC) about Heisenberg have you? 8^) The users of computers in the world have to separate economic competitiveness (contest) from truth. Optimization is a short cut. That's valid. Using short-cuts for some contests is "cheating." It can be a fine line (few real cheats yet). >Something like the bc or nbc benchmarks may be not >very good, but they are cheap to run. DO not be lulled by the dark side of the Force. Luke.... Don't start bad habits. Do not be suckered by cheap wine. 8^) >Does SPEC alone allow one to characterize the performance of a system? >Definately not. Does it help? Sure. How about TPC-A? For any single >characterization, one can cite exceptions. Only the complete universe >of information is universally useful. Simply: no the complete universe is not necessary. I can give examples from science. One wants to know the key basis of information: rates, constants, anomalies (gottchas), exceptions, etc. >Performance information is damaging only if it is missued (happends a lot). There is a tendency in organizations to kill the bearer of bad news. This is an unfortunate trend because many aspects of research are perceived to involve failure. And when you are pushing the frontiers of knowledge and information building faster machines it doesn't encourage development. It is the attitude of these organizations which have to change. Anyway, on a slightly different note: those people reading this from companies where I have spent a lot of time. Our benchmarking concluded last evening at 830pm. We thank those who helped us: Alan, Richard, Keith, Ken, Olivier (and Linda as lunch company), Greg, Sundar, Gwen, Vicki, Ron, John, Bill, Steve, Judy, and tons of others (wow late evenings...). And a note: (SPEC members take note) Tom (Woodrow) discovered a bug in MUSBUS/KENBUS: In getwork.c an initialization is necessary: near line 150: add the line mentioned below: .. /* gobble input */ scrp->buf = (char *)malloc(512); scrp->blen = 0; /* add this line */ while ((i = read(f, &scrp->buf[scrp->blen], 512)) > 0) { .. Back to commenting on benchmarking when I get back from yet another trip to Utah (January). Merry Christmas, Happy New Year. --e.n. miya, NASA Ames Research Center, eugene@eos.arc.nasa.gov {uunet,mailrus,most gateways}!ames!eugene AMERICA: CHANGE IT OR LOSE IT.