benchmark

Benchmarking on the Cray YMPhttp://people.sc.fsu.edu/jburkardt/presentations/benchmark.pdf..........John Burkardt08 March 2008AbstractThis report discusses the idea of benchmarking programs on the CrayYMP at the Pittsburgh Supercomputing Center. It was written around1990. The machines and programs discussed here are long gone, but thematerial is not merely of nostalgic interest.1 IntroductionBenchmarking implies comparison. We might want to compare: two programs, two versions of a program, two compilers, two machines.But remember that you can’t compare one thing. It’s actually hard to knowwhat to make of the fact that your program runs at a rate of 87 MegaFLOPS, orcompletes in 17 seconds; it’s much easier to understand that your program runstwice as fast on an Amiga as it does on a Commodore 64, or that the Pascalversion takes 10 times as long to run as the PL/1 version.After all the theoretical work that goes into writing a program, benchmark-ing is the practical side. No matter how well the program ought to run, bench-marking reports what really happened. It can be a useful (and sometimesshocking) guide to how well you are doing in matching your programming towhat the machine does well.We can’t make benchmarking an end in itself. You’re here to get answers,and the only role of benc is to show you whether you’re getting thoseanswers quickly and e ciently. Use benchmarking to determine whether thereare any serious performance failures in ...