On 4/17/2013 6:41 PM, José Orlando Pereira wrote: > On Wednesday 17 April 2013 17:38:40 Fabio M. Di Nitto wrote: >> >> The patch is better. >> >> Did you verify the performance impact of switching from static to malloc? >> >> there is a test/cpgbench you can use. > > Actually cpgbench doesn't show the overhead, as it uses CS_DISPATCH_BLOCKING. After > changing the following in cpgbench.c: > > - cpg_dispatch (handle, CS_DISPATCH_BLOCKING); > + while(1) cpg_dispatch (handle, CS_DISPATCH_ONE); > > I'm getting mixed results (worse for 64 bytes, better but unstable for 8k > and 40k). I'll investigate further and let you know. Ok, try the basic: CS_DISPATCH_ONE with|without your patch and see if you notice the same mixed results (might be that we need to adapt the test suite to measure your patch). Fabio _______________________________________________ discuss mailing list discuss@xxxxxxxxxxxx http://lists.corosync.org/mailman/listinfo/discuss
