Instantaneous snapshot of CPU schedule

[J K Rai]

Hello

Is there any utility using which we can take current (instantaneous) snapshot of process schedule on the linux based system from user land. Like use a system call or command to know what is running on each of the cpus 0..n-1 (assuming n cpus in the system). Output could be like tuple (pid, cpu_no).

If such utility does not exist then kindly throw some light / ideas so that it could be implemented.

Regards,
Jitendra


      Add whatever you love to the Yahoo! India homepage. Try now! http://in.yahoo.com/trynew
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[Alan Cox]

On Tue, 27 Oct 2009 13:14:34 +0530 (IST)
J K Rai <jk.anurag@...> wrote:

> Hello
>
> Is there any utility using which we can take current (instantaneous) snapshot of process schedule on the linux based system from user land. Like use a system call or command to know what is running on each of the cpus 0..n-1 (assuming n cpus in the system). Output could be like tuple (pid, cpu_no).
>
> If such utility does not exist then kindly throw some light / ideas so that it could be implemented.

I'm not sure it would be terribly useful. You know what is running on at
least one of the processors - your snapshot tool. So for uniprocessor its
100% useless, for dual processor 50% useless.

It would also be incredibly expensive to serialize everything on the
system just to take a measurement.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

[Ingo Molnar]

* J K Rai <jk.anurag@...> wrote:

> Hello
>
> Is there any utility using which we can take current (instantaneous)
> snapshot of process schedule on the linux based system from user land.
> Like use a system call or command to know what is running on each of
> the cpus 0..n-1 (assuming n cpus in the system). Output could be like
> tuple (pid, cpu_no).
>
> If such utility does not exist then kindly throw some light / ideas so
> that it could be implemented.

We have a tool that can record scheduling events and analyze/display
them, see:

  perf sched record

  perf sched latency
  perf sched map
  perf sched trace
  perf sched replay

the latest version is in:

  http://people.redhat.com/mingo/tip.git/README

do 'cd tools/perf; make -j install' to get the 'perf' command.

Now, your specific question was 'instantaneous'. I wouldnt mind such an
extension to 'perf sched record':

  perf sched record --snapshot

or so. Then 'perf sched map' would show which task is running on which
CPU. Other output modes would be possible too.

Would you be interested in extending 'perf sched' in such a fashion?

        Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/