Greg KH wrote: >A better benchmark is lmbench for micro results > Agreed: lmbench is a terrific piece of science & engineering, meticulous microbenchmarking. > and the contest test for >macro ones: > http://members.optusnet.com.au/ckolivas/contest/ >which a lot of people are using these days to test out mm changes to the >kernel. > That's interesting. There is a need for a good, standard, open source system macrobenchmark. Lots of people (WireX included) use various forms of "build the kernel from source" as a benchmark, but because the kernel changes, the benchmark is only useful for relative comparisons, i.e. I can use it to show that the Foo feature imposes 4% overhead, but you can't go off and run another benchmark to compare your results to mine. So it is a very good thing that someone is working to standardize it. On the other hand, reading the Contest web site, they are a bit vague on some standards. The primary glaring example is that they are non-specific about kernel versions, just suggesting that you get one, or default to 2.4.19. To be a useful standard benchmark, they would need to freeze on a particular kernel version (it is just a workload for the benchmark). Still, it's good to see work going into standardizing macro benchmarks. Caveat: there is a standard system throughput macrobenchark from SPEC, but it is not open source, and is in fact quite expensive. Crispin -- Crispin Cowan, Ph.D. Chief Scientist, WireX http://wirex.com/~crispin/ Security Hardened Linux Distribution: http://immunix.org Available for purchase: http://wirex.com/Products/Immunix/purchase.html Just say ".Nyet"
This archive was generated by hypermail 2b30 : Fri Dec 27 2002 - 18:17:43 PST