eugene@pioneer.arpa (Eugene Miya N.) (01/12/88)
Last week, I had an idea which I'm adding to a paper (the notorious paper) I'm writing. The idea concerns itself with the conditions under which benchmarking is done. The grading is not directly in the measurement scope of a benchmark itself (This, I think is best covered by David Bailey and John Barton in the NAS Kernel benchmarks with their 0-1000 lines of modification). The grading for measurement conditions is as follows: E: Measurement program run on a timeshared system with no controls what so ever. (Assumes no logging of system load or other measure.) This condition is perfectly fine for most development work. D: Measurement program with restricted limited usage (may be some background work). C: Conventional "stand-alone" measurement. Can be a single-user system. This is where most benchmarks fall, and it's conveniently in the middle as well. B: "Stand-alone" measurement with some controls: daemons either turned off or killed. A: Extremely elaborate measurement and recording facilities (hardware as well as software), high precision clock and instruction monitoring, ability to control environment using direct control (using commands) AND indirect control using controlled work loads. Comments by mail please. From the Rock of Ages Home for Retired Hackers: --eugene miya, NASA Ames Research Center, eugene@ames-aurora.ARPA "You trust the `reply' command with all those different mailers out there?" "Send mail, avoid follow-ups. If enough, I'll summarize." {uunet,hplabs,hao,ihnp4,decwrl,allegra,tektronix}!ames!aurora!eugene