libvirt users - Feb 2011 - monitiring cpu usage via cgroup

Hi

I was asking about the fedora 14 kernel if it is good enough for cgroup 
usage because 
I am trying to set a cgroup  under  cpu subsytem ( /dev/cgroup/cpu/group1/ 
) that have   /cpu.rt_runtime_us of 100000
while cpu.rt_period_us has a value of 1000000 i.e a ratio of 1/10 . still 
when I run a task (endless loop) in that group
(cgexec -g cpu,cpuset:group1 ./test)  it gets all the cpu core time that 
is assigned to it ( i watch via top utility)  , so it seems
that the quota set via the group does not take effect although I restart 
the cgconfig service .
I also verified   libcgroup is installed by:  rpm -q libcgroup

Am I missing something or top is not the right utility to  watch the cpu 
usage in this case ?

thanks

Zvi Dubitzky 
Email:dubi at il.ibm.com

On Tue, 15 Feb 2011 09:52:51 +0200
Zvi Dubitzky <DUBI at il.ibm.com> wrote:
> Hi
> 
> I was asking about the fedora 14 kernel if it is good enough for cgroup 
> usage because 
> I am trying to set a cgroup  under  cpu subsytem ( /dev/cgroup/cpu/group1/ 
> ) that have   /cpu.rt_runtime_us of 100000
> while cpu.rt_period_us has a value of 1000000 i.e a ratio of 1/10 . still 
> when I run a task (endless loop) in that group
> (cgexec -g cpu,cpuset:group1 ./test)  it gets all the cpu core time that 
> is assigned to it ( i watch via top utility)  , so it seems
> that the quota set via the group does not take effect although I restart 
> the cgconfig service .
> I also verified   libcgroup is installed by:  rpm -q libcgroup
> 
> Am I missing something or top is not the right utility to  watch the cpu 
> usage in this case ?
> 
rt_runtime_us and rt_period_us are parameters for handling realtime class
(SCHED_RR, SCHED_FIFO) applications. Does your 'test' run in which
priority ?
(see sched_setscheduler())


Thanks,
-Kame
> thanks
> 
> Zvi Dubitzky 
> Email:dubi at il.ibm.com
> 
> 
> _______________________________________________
> libvirt-users mailing list
> libvirt-users at redhat.com
> https://www.redhat.com/mailman/listinfo/libvirt-users
>

thanks Kame,

1) the policy of my process(test) is indeed SCHED_NORMAL .
     When I use sched_setscheduler() to set the pid sched policy  to 
SCHED_FIFO then the cpu usage via cgroup  rt_runtime_us and 
  rt_period_us  is affected . Still  the ratio I specified  is 1:10   but 
I see via top that my process 
     ,that gets 1 cpu ,shows 20% usage

2)  What about memory and cpuset subsystems of cgroup. Do their settings 
apply also only in the cases of processes with policy       SCHED_FIFO/SCHED_RR
?

3) lastly:
     What is the meaning of the ?cap? parameter in the schedinfo virsh 
command (what are the units and range of values ) and what is 
    the ?weigth? parameter .I did not find any documentation of libvirt 
about it ?

thanks

Zvi Dubitzky 
Email:dubi at il.ibm.com