Carlo said: > Martin said: >> There is a fix in linux/metrics.c r1029 that bz#180 . This should go >into 3.0.8. The fix is relatively nice in trunk, but more ugly in 3.0.x due >to the code duplication in the networking metrics. I see two possible >ways: >> >> a) we backport only r1029, even if if it is ugly > >probably better to minimize the risk, there are other changes as well which >could be cherry-pick into stable independently, like r860, or some that >might need more testing like r1008.
I like r860 and r1008 is #ifdef'd off by default. It looks reasonable for my hosts though, and so I am tempted to enable it to go ahead and get rid of my petabyte/second spikes in 3.0. >> b) we take the current state of linux/metrics.c minus r1010 /float >conversion for memory metrics) This is the option I am currently testing on a couple hosts and am hoping to deploy in production. Seems to be working well, but I haven't had it running long or on many hosts. >this will pull also the fix in r860, but also other potentially >destabilizing >patches which are not critical bugfixes, like r986, r1013, r1146, r1161 and >r1162. It is hard to say whether they are needed in stable. But I am tempted to give it all a try. I like the networking rewrite; thanks Martin! I agree with your FIXME comments that similar rewrites are needed where other files are read multiple times. I occasionally see 100+% spikes in my CPU graphs and I think if we did this for /proc/stat, then we could fix this too, as documented in bz#131. -twitham ------------------------------------------------------------------------- This SF.net email is sponsored by the 2008 JavaOne(SM) Conference Don't miss this year's exciting event. There's still time to save $100. Use priority code J8TL2D2. http://ad.doubleclick.net/clk;198757673;13503038;p?http://java.sun.com/javaone _______________________________________________ Ganglia-developers mailing list Ganglia-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/ganglia-developers