[RFC][PATCH 0/6] Add group fairness to CFS - v1

Srivatsa Vaddagiri vatsa at linux.vnet.ibm.com
Fri Jun 15 07:06:45 PDT 2007


On Fri, Jun 15, 2007 at 04:46:13PM +0400, Kirill Korotaev wrote:
> Ingo Molnar wrote:
> 
> > i'd still like to hear back from Kirill & co whether this framework is 
> > flexible enough for their work (OpenVZ, etc.) too.
> 
> My IMHO is that so far the proposed group scheduler doesn't look ready/suitable.

Hi Kirill,
	Yes its work-in-progress and hence is not ready/fully-functional
(yet). The patches I posted last gives an idea of the direction it is
heading. For ex: http://lkml.org/lkml/2007/6/11/162 and
http://lkml.org/lkml/2007/5/25/146 gives an idea of how SMP load balance will 
works.

IMHO the nice thing about this approach is it (re)uses lot of code in
scheduler which is there already to achieve fairness between tasks and
higher schedulable elements (users/containers etc).

Also with CFS engine's precise nanosecond accurate accounting and time-sorted 
list of tasks/entities, I feel we will get much tighter control over 
distribution of CPU between tasks/users/containers.

> We need to have a working SMP version before it will be clear
> whether the whole approach is good and works correct on variety of load patterns.

If you have any headsup thoughts on areas/workloads where this may pose problems
for container/user scheduling, I would be glad to hear them. Otherwise I would 
greatly wellcome any help in developing/reviewing these patches which meets 
both our goals!

-- 
Regards,
vatsa


More information about the Containers mailing list