[Devel] Re: [PATCH -mm 5/5] swapcgroup (v3): implement force_empty

Daisuke Nishimura nishimura at mxp.nes.nec.co.jp
Fri Jul 4 05:33:01 PDT 2008


On Fri, 4 Jul 2008 19:16:38 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu at jp.fujitsu.com> wrote:
> On Fri, 4 Jul 2008 15:24:23 +0900
> Daisuke Nishimura <nishimura at mxp.nes.nec.co.jp> wrote:
> 
> > This patch implements force_empty of swapcgroup.
> > 
> > Currently, it simply uncharges all the charges from the group.
> > 
> > I think there can be other implementations.
> > 
> > What I thought are:
> > - move all the charges to its parent.
> > - unuse(swap in) all the swap charged to the group.
> > 
> 3. move all swap back to memory (see swapoff.)
> 
> 
Do you mean swapping in all the swap including used by
other groups?
It would be one choice anyway.

> > But in any case, I think before implementing this way,
> > hierarchy and move charges support are needed.
> > 
> > So I think this is enough for the first step.
> > 
> 
> I don't think hierarchy/automatic-load-balancer for swap cg is necessary.
It's the problem of how the "hierarchy" would be, I think.
I'm saying "hierarchy" here just to mean "some kind of feature
where a parent includes their children".
I think "hierarchy" is needed if we implement the choice 1 above,
and I personally think it would be the best choice.

> Hmm...but handling limit_change (at least, returns -EBUSY) will be necessary.
I think so too.
But I'm not sure now it's good or bad to support shrinking at limit_change
about swap.
Shrinking swap means increasing the memory usage and that may cause
another swapout.

> Do you consider a some magical way to move pages in swap back to memory ?
> 
In this patch, I modified the find_next_to_unuse() to find
the entry charged to a specific group.
It might be possible to modify try_to_unuse()(or define another function
based on try_to_unuse()) to reduce swap usage of a specified group
down to some threashold.
But, I think, one problem here is from which device the swaps
should be back to memory, or usage balance between swap devices.

> In general, I like this set but we can't change the limit on demand. (maybe)
> (just putting it to TO-DO-List is okay to me.)
> 
I'm sorry but what do you mean by "change the limit on demand"?
Could you explain more?


Thanks,
Daisuke Nishimura.
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers




More information about the Devel mailing list