Re: [PATCH] KVM simplified virtio balloon driver

[Rusty Russell <rusty-8n+1lVoiYb80n/F98K4Iww@public.gmane.org>]

On Thursday 17 January 2008 15:01:46 Anthony Liguori wrote:
> Rusty Russell wrote:
> > OTOH it's currently pretty obvious (and usually fatal) if the guest has
> > trouble meeting the balloon requirements.  A serious host needs a way of
> > detecting stress in the guest anyway, which this doesn't offer until it's
> > too late...
>
> The question I'm interested in answering though is not if but when.  I
> would like to know when the guest has reached it's target.

I'm saying that it will be v. quickly in all but "too much squeeze" case.

> And while we do get the madvise call outs, it's possible that pages have
> been faulted in since then.

But that's exactly what the balloon number *doesn't* tell you.  It can tell 
you that it's released pages back to be used by the OS, but not whether the 
OS has used them.

I think this number is good for debugging the balloon driver, but for anything 
else it's a false friend.

Rusty.
PS.  Please cut down mails when you reply.

-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/