My understanding of ruby-cycles is that it is basically the global-time on the event-queue when the dump-stats is triggered. But some of my results seem really weird even on the base MOSI_SMP_bcast. Using a hierarchical tree switch. I am running a shared counter protected by TATAS. Compared to execution on a real sunfire U6800. There is significant degradation in the benchmark performance. Especially the numbers for the Ruby-cycles seem to be N_{procs}*T_{expected}. Is there something I am missing ?
New Yahoo! Messenger with Voice. Call regular phones from your PC for low, low rates.
|