bdg@tetons.UUCP (Blaine Gaither) (09/29/90)
The November Issue of ACM Sigmetrics Performance Evaluation Review will feature an editorial on Commercial Benchmark Suites. Of particular interest are the Neal Nelson benchmark, the UnixWorld/NealNelson benchmark, AIM benchmarks or any others that our readership wishes to discuss. Please address your comments to the editor. The editorial was already partly complete before the UnixWorld article (discussed in the August issue of PER) was published. The UnixWorld article seemed particularly irresponsible. The damage caused by a possibly flawed benchmark is not limited to misleading the benchmark's clients, but also tends to cause redirection of vendor resources into areas which may not give maximum gain to the customers. Even IBM has scarce resources. At this stage in the RS/6000 development, is it constructive to "tune" awk as opposed to other system characteristics? From the August issue: Coming Editorial - Call for letters-to-the-editor AWK THE NEAL NELSON BENCHMARK OR UNIXWORLD'S SHAME By: Blaine Gaither The September 1990 issue of UnixWorld just crossed my desk. On page 13 in a section titled "INDUSTRY NEWS" there is an article titled "BIG BLUE'S RS/6000 ISN'T WHITE LIGHTNING". In this article it is stated that recent UnixWorld/Neal Nelson & Associates benchmarking of the RS/6000 POWERserver 320 found, among other things, that the RS/6000 was about 20% slower than the Hewlett-Packard Vectra 486 in floating- point and integer calculations. The article goes on to quote Nelson(sic) as offering two possible explanations for the poor floating-point results: 1) IBM could have a slow implementation of awk, which is used in the floating-point tests; or 2) The IBM system's floating-point speed is less than they expected. I have another possible explanation: Could the UnixWorld/Neal Nelson benchmark have a poor experimental design and give erroneous results? Should the title of the article have been: "The IBM RS/6000 Shows the UnixWorld/Neal Nelson Benchmark to be Flawed"? For those of you who aren't familiar with awk, it is a utility for pattern scanning and processing. It is often implemented as an interpreter and it does its internal arithmetic in floating-point. Criticism of the Neal Nelson & Associates benchmarking methodology has been floating around the performance community for years. The next issue of PER will feature an editorial on this and other commercial benchmarking suites. If you have comments you wish to publish on these topics please send a letter to the editor by November 7. Standard disclaimers. These are my opinions, not my employers or the ACM's. Should any opinions or data published as fact prove incorrect I will gladly publish a correction. Blaine Gaither Amdahl Corporation 143 No. 2nd East St., Rexburg, Idaho 83440-1619 UUCP:{ames,decwrl,sun,uunet}!amdahl!tetons!bdg (208) 356-8915 INTERNET: bdg@tetons.idaho.amdahl.com