home *** CD-ROM | disk | FTP | other *** search
- Newsgroups: comp.benchmarks
- Path: sparky!uunet!spool.mu.edu!agate!ames!data.nas.nasa.gov!amelia.nas.nasa.gov!eugene
- From: eugene@amelia.nas.nasa.gov (Eugene N. Miya)
- Subject: [l/m 3/17/91] Ways to fool the masses with benchmarks (15/28) c.be. FAQ
- Keywords: who, what, where, when, why, how
- Sender: news@nas.nasa.gov (News Administrator)
- Organization: NAS Program, NASA Ames Research Center, Moffett Field, CA
- Date: Tue, 15 Dec 92 12:25:10 GMT
- Message-ID: <1992Dec15.122510.21215@nas.nasa.gov>
- Reply-To: eugene@amelia.nas.nasa.gov (Eugene N. Miya)
- Lines: 109
-
- 15 12 Ways to Fool the Masses with Benchmarks <This panel>
- 16 SPEC
- 17 Benchmark invalidation methods
- 18
- 19 WPI Benchmark
- 20 Equivalence
- 21 TPC
- 22
- 23
- 24
- 25 Ridiculously short benchmarks
- 26 Other miscellaneous benchmarks
- 27
- 28 References
- 1 Introduction to the FAQ chain and netiquette
- 2
- 3 PERFECT Club
- 4
- 5 Performance Metrics
- 6
- 7 Music to benchmark by
- 8 Benchmark types
- 9 Linpack
- 10
- 11 NIST source and .orgs
- 12 Measurement Environments
- 13 SLALOM
- 14
-
- From David Bailey et al.
-
- Quote only 32-bit performance results, not 64-bit results.
-
- Present performance figures for an inner kernel as the performance of the
- entire application.
-
- Quietly employ assembly code and other low-level language constructs.
-
- Scale up the problem size with the number of processors, but omit any
- mention of this fact.
-
- Quote performance results projected to a full system.
-
- Compare your results against scalar, unoptimized code on Crays.
-
- When direct run time comparisons are required, compare with an old code
- on an obsolete system.
-
- If megaFLOPS rates must be quoted, base the operation count on the parallel
- implementation, not on the best sequential implementation.
-
- Quote performance in terms of processor utilization, parallel speedup or
- megaFLOPS per dollar.
-
- Multilate the algorithm used in the parallel implementation to match
- the architecture.
-
- Measure parallel run times on a dedicated system, but measure conventional
- run times in a busy environment.
-
- If all else fails, show pretty pictures and animated videos, and don't talk
- about performance.
-
- Ref. 1
- NAS TR #
-
- Ref. 2
- %A David Bailey
- %T Twelve Ways to Fool the Masses when Giving Performance Results on
- Parallel Computers
- %J Supercomputing Review
- %V 4
- %N 8
- %D August 1991
- %P 54-55
-
- References
- Darrell Huff, How to Lie with Statistics
- How to Lie with Maps
-
- Gordon Bell's 11 rules of supercomputer design:
-
- 1) Performance, performance, performance.
- 2) Everything matters.
- 3) Scalar matter most.
- 4) Provide vectors as price allows.
- 5) Avoid holes in performance.
- 6) Place peaks in performance.
-
- 7) Provide a decade of addressing.
- 8) Make it easy to use.
- 9) Build on others work.
- 10) Design for the next one (and do it again) [three times]
- 11) Have slack resources.
-
- ^ A
- s / \ r
- m / \ c
- h / \ h
- t / \ i
- i / \ t
- r / \ e
- o / \ c
- g / \ t
- l / \ u
- A / \ r
- <_____________________> e
- Language
-
-