Re: [PATCH 23/24] cfq-iosched: collect stats from dead cfqgs

[Tejun Heo <tj-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

Hey, Vivek.

On Wed, Jan 02, 2013 at 11:44:15AM -0500, Vivek Goyal wrote:
> > > In fact if we have a realiable way of resetting status then online/offline
> > > infrastructure might not be required? I think per cpu stats will be a
> > > problem though and that's why we probably require logic to online/offline
> > > the group?
> > 
> > Hmmm?  What do you mean?
> 
> I mean if we had a reliable way of resetting stats after transferring
> then we would not need to keep a track of whether group is online/offline.
> We could add everything and adding zero will not change anything. In fact
> it will also take care of residual IO (IO which happened after transfer
> of stats).

Ah... yeah, if we can do atomic transfer, we might be able to do away
with on/offlining.  Couldn't think of a way to do that without
incurring overhead to hot paths.

> Or I missed the real reason of why do we have group online/offline
> infrastructure.

But given that on/offline state is something common to cgroup, I don't
think adding the states to blkcg is a bad idea.  We need it for
reliable iterations anyways.

Thanks.

-- 
tejun