Re: [PATCH mmotm 2/2] memcg: allow mem.limit bigger thanmemsw.limit iff unlimited

[Daisuke Nishimura]

On Wed, 3 Jun 2009 17:46:41 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> On Wed, 3 Jun 2009 14:01:02 +0900
> Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> wrote:
> 
> > On Wed, 3 Jun 2009 12:52:28 +0900, KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> > > On Wed, 3 Jun 2009 11:50:27 +0900
> > > Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> wrote:
> > > 
> > > > Now users cannot set mem.limit bigger than memsw.limit.
> > > > This patch allows mem.limit bigger than memsw.limit iff mem.limit==unlimited.
> > > > 
> > > > By this, users can set memsw.limit without setting mem.limit.
> > > > I think it's usefull if users want to limit memsw only.
> > > > They must set mem.limit first and memsw.limit to the same value now for this purpose.
> > > > They can save the first step by this patch.
> > > > 
> > > 
> > > I don't like this. No benefits to users.
> > > The user should know when they set memsw.limit they have to set memory.limit.
> > > This just complicates things.
> > > 
> > Hmm, I think there is a user who cares only limitting logical memory(mem+swap),
> > not physical memory, and wants kswapd to reclaim physical memory when congested. 
> > At least, I'm a such user.
> > 
> > Do you disagree even if I add a file like "memory.allow_limit_memsw_only" ?
> > 
> 
> How about removing memory.limit < memsw.limit condition completely ?
> 
It might be good idea.
IMHO, there is no critical reason it must be checked by kernel, but I'm not sure.

All I wanted to do was "let users who cares only about memsw.limit
ignore mem.limit completely". That's why, I treated only the "unlimited"(not set
mem.limit) case as special.
But, as you say, there is no reason it must be implemented in kernel.
(We can use a middle-ware or something.)

I'll drop this and consider more.


Thanks,
Daisuke Nishimura.
--
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/