Re: [ckrm-tech] [RFC][PATCH] UBC: user resource beancounters

From: Chandra Seetharaman
Date: Mon Aug 21 2006 - 17:02:03 EST


On Mon, 2006-08-21 at 14:55 +0400, Kirill Korotaev wrote:
<snip>

> >>If you have a single container controlling all the resources, then
> >>placing kjournald into CPU container would require setting
> >>it's memory limits etc. And kjournald will start to be accounted separately,
> >
> >
> > Not necessarily. You could just set the CPU shares of the group and
> > leave the other resources as don't care.
> don't care IMHO doesn't mean "accounted and limited as container X".
> it sounds like "no limits" for me.

Yes. But, it would provide the same functionality that you want (i.e
limit only CPU and no other resources).

>
> >>while my intention is kjournald to be accounted as the host system.
> >>I only want to _guarentee_ some CPU to it.
> > I do not see any _guarantee_ support, only barrier(soft limit) and
> > limit. May be I overlooked. Can you tell me how guarantee is achieved
> > with UBC.
> we just provide additional parameters like oomguarpages, where barrier
> is a guarantee.

I take it that you are suggesting that the controller can use barrier as
guarantee.

I don't see how it will work. charge_beancounter() returns -ENOMEM even
when the group is over its barrier (when queried with strict ==
UB_BARRIER).

I have to see the oomguarpatches patches for understanding this, I
suppose.
>
> Kirill
--

----------------------------------------------------------------------
Chandra Seetharaman | Be careful what you choose....
- sekharan@xxxxxxxxxx | .......you may get it.
----------------------------------------------------------------------


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