On 12/08/2013 04:43 PM, Devon H. O'Dell wrote: >> This is a simple `perf top -p $PID` on one of of our gmetad nodes >> > >> >Samples: 1M of event 'cycles', Event count (approx.): 64115959770 >> > 6.59% libexpat.so.1.5.2 [.] 0x0000000000011b8d >> > 4.77% libganglia-3.6.0.so.0.0.0 [.] hashval >> > 2.62% [kernel] [k] __d_lookup >> > 2.21% [kernel] [k] _spin_lock >> > 2.14% libc-2.12.so [.] vfprintf >> > 1.61% librrd.so.4.2.0 [.] process_arg >> > 1.54% libganglia-3.6.0.so.0.0.0 [.] hash_lookup >> > 1.46% [kernel] [k] __link_path_walk >> > 1.16% libc-2.12.so [.] __GI_____strtod_l_internal >> > 1.11% libc-2.12.so [.] memcpy >> > 1.08% libc-2.12.so [.] _int_malloc >> > >> >So I suppose my intuition about xml parsing expense is off. I have not used >> >perf as much as I should, if we were seeing similar rrd writing contention >> >should I literally see "stat" near the top? > Ah, so to see what's really going on: > > perf record -e cpu-clock -g -p $PID > > Let that run for a minute or two. Then: > > perf report --sort=comm,dso,symbol -G > > If you don't have cpu-clock, cycles is OK, but you definitely are > going to want to see the callgraph. The time in XML is mostly writing > RRDs and you only see that digging down into the chain. >
For the list, Devon and I spoke in #ganglia and the high occurrence of libexpat in this sample seems to be an artifact of missing debug symbols. ------------------------------------------------------------------------------ Rapidly troubleshoot problems before they affect your business. Most IT organizations don't have a clear picture of how application performance affects their revenue. With AppDynamics, you get 100% visibility into your Java,.NET, & PHP application. Start your 15-day FREE TRIAL of AppDynamics Pro! http://pubads.g.doubleclick.net/gampad/clk?id=84349831&iu=/4140/ostg.clktrk _______________________________________________ Ganglia-developers mailing list Ganglia-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-developers
