Hello!
I see load average value changed (decreased) comparing to CentOS5. For example - I have 4 servers, 3 under CenOS5, and 4th under CentOS6, the loads divided using nginx equally, hardware are the same, using top, htop, munin I see CPU loads, network loads, iowaits are absolutely the same, but LA shows ~5.5 on each CentOS5 servers and ~1.5 on CentOS6 machine. And recently I reinstalled one more server from CentOS5 to CentOS6, and LA value decreased also from 5 to 1.
It's not a big problem, but I can't understand why... because I see the loads did not changed, 8 CPUs server have 4 CPUs at ~100% all the time, but LA now shows ~1-1.5 instead of 5-5.5, that I saw using CentOS5. Did LA weights changed in Centos6 or something?
Regards,
renfry
CentOS 6 load average value (LA)
Re: CentOS 6 load average value (LA)
here is Munin pictures of one more CentOS 6 server:
http://xmages.net/i/3249162
it is 2xXeon E5405, 8 CPUs server, here CPU and LA usage in attachment. You may see that I use 2-3 CPUs, no iowaits, using CentOS 5 I saw LA 3-4, but now I see 0.8-1.5, it is strange.
For example it is usual for me to see one 100% CPU core usage is 1 LA, but now it looks strange.
http://xmages.net/i/3249162
it is 2xXeon E5405, 8 CPUs server, here CPU and LA usage in attachment. You may see that I use 2-3 CPUs, no iowaits, using CentOS 5 I saw LA 3-4, but now I see 0.8-1.5, it is strange.
For example it is usual for me to see one 100% CPU core usage is 1 LA, but now it looks strange.
Re: CentOS 6 load average value (LA)
I've seen the same problem; I believe that the kernel has changed how it calculates load average and no longer includes threads as part of the calculation. For example, I had a processor that had multiple cores reporting 50% utilization in "top" but if you looked at the process state reported in top it said all of the processes were "interruptible sleep". Furthermore top also only reported "1 running task" which was top itself. If I switched "top" into threads mode the running tasks number went up to 9-17 and I could see the threads in the "R" state.
I'm not sure if it was intentional to change how loadavg was calculated or if it was inadvertent due to changes in the scheduler. If loadavg no longer includes the effect of threads...then it isn't a very useful metric.
I'm not sure if it was intentional to change how loadavg was calculated or if it was inadvertent due to changes in the scheduler. If loadavg no longer includes the effect of threads...then it isn't a very useful metric.
Re: CentOS 6 load average value (LA)
So after some more research, this is probably due to various bugs reporting loadavg on a tickless (NO_HZ=y) kernel.
http://www.gossamer-threads.com/lists/linux/kernel/1487817
http://www.gossamer-threads.com/lists/linux/kernel/1487817