Hi guys,
I have a new kvm cgroup test for you.
This time it's in cpu cgroup and it uses cpu.shares capability. It modifies
the number of shares/time_slices which each process gets from the scheduler
As the verification is kind of tricky I'm testing only three different
scenerios and only symetrical solutions:
1) #threads == #CPUs
2) #threads + 1 == #CPUs, +1thread have the lowest priority (or equal)
3) #threads * #cgroups == #CPUs
ad1) When less threads are occupied, the utilisation should be equal.
ad2) When there is +1 thread with lower priority, it fight with the
second thread from the lowest priority group, thus the utilisations
are still equal. If you try it other way around (higher utilisation)
this current thread forces the utilisation in cgroups ratios. This
is not tested as it would be trickier with multiple cgroups/threads
ad3) When each cgroup have multiple threads it can fight for it's share
thus in the end they all should settle on cgroup ratio utilisations.
To minimalize system impact I'm using 10-100times bigger slices/priority
then other processes. Still when are you running other processes on the
system the test might fail. Keep in mind it should run on clean test
machine.
Anyway, be my guests at github pull request:
https://github.com/autotest/autotest/pull/94
Regards,
Lukáš Doktor
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
