[netperf-talk] CPU utilization & tickless kernels?

[Andrew Gallatin]

On 12/07/12 17:26, Rick Jones wrote:

> Ostensibly, the 2.6.38 kernel used above is tickless:
> 
> raj at tardy:~$ uname -a
> Linux tardy 2.6.38-16-generic #67-Ubuntu SMP Thu Sep 6 17:58:38 UTC 2012
> x86_64 x86_64 x86_64 GNU/Linux
> raj at tardy:~$ grep HZ /boot/config-2.6.38-16-generic
> CONFIG_RCU_FAST_NO_HZ=y
> CONFIG_NO_HZ=y
> CONFIG_HZ_100=y
> # CONFIG_HZ_250 is not set
> # CONFIG_HZ_300 is not set
> # CONFIG_HZ_1000 is not set
> CONFIG_HZ=100
> CONFIG_MACHZ_WDT=m
> 
> and while my investigation wasn't exhaustive, my tests like those above
> seemed to show agreement between netperf and vmstat - it was showing
> about 79% idle in the overall.
> 

FWIW, my issue was on Ubuntu with their 3.5.0-19-generic.  It took
longer than I wanted, since ubuntu no longer offers separate server
kernels with the tickless stuff turned off. So I took the latest
kernel.org 3.5.7, and used the Ubuntu "generic" config.  The resulting
kernel behaved the same way.  I then made this change to the config &
rebuilt, and the problems went away:

diff config.ub .config
114c114
< CONFIG_NO_HZ=y
---
> # CONFIG_NO_HZ is not set
125d124
< CONFIG_RCU_FAST_NO_HZ=y
589d587
< CONFIG_CPU_IDLE_GOV_MENU=y
3219a3218
> # CONFIG_NTP_PPS is not set

Interestingly, the usage goes up to ~80%, which is higher than either
the vmstat or netperf calculated averages on the tickless kernels.

At any rate, this looks like a core linux kernel issue on this hardware.
I guess I'll try the latest 3.7 rc at some point.  But given that the
"summary" and individual counters are apparently both wrong, it does
not seem like this is a netperf issue in any way.  So, sorry for
the noise..

Drew