 |
 |
On Thu, 10 Sep 2009, Burke, Stephen (STFC,RAL,PPD) wrote: > LHC Computer Grid - Rollout >> [mailto:[log in to unmask]] On Behalf Of Mattias Wadenstein > said: >> The hep-spec benchmark is per node, at least that the number >> the official benchmark script outputs (which automatically runs one > spec >> 2006 all_cpp >> instance per cpu reported in /proc/cpuinfo - including >> hyperthreading- > "CPU:s"). > > That's true, but if you look at the document (note 1.1d on page 6) it > says "The rule established by WLCG to publish the correct value for > BenchmarkSI00 is described in the following GOC wiki page (please check > in particular the example given): > http://goc.grid.sinica.edu.tw/gocwiki/CPU_Benchmarking_HowTo. Please, > note that in this wiki page one is instructed to find the power per job > slot (dividing the total score from the HEPIX procedure by the number of > job slots). The proposal we make is to find the power per core (dividing > the total score by the number of cores in the box). Ah, totally new documents to me. Good thing I happened to be reading this thread, because I'm just implementing the re-publishing of ARC infosys into the glite BDII. Wouldn't it make more sense to run the all_cpp benchmarks in the same number of copies as the job slots on the node though? Especially if one would use those numbers for brokering, which I think is one of the intended uses. /Mattias Wadenstein