RE: [Xen-devel] [PATCH] Paging and memory sharing for HVM guests

[Dan Magenheimer <dan.magenheimer@xxxxxxxxxx>]

> (albeit slow if many pages have been swapped out)

That's one thing I'm worried about.  Even if the migration
eventually completes, it is my understanding that "bad
things" can happen if a live migration takes too long.

Another thing I'm worried about is a deadlock (or maybe
livelock) if memory is overcommitted, a live migration
is called for of a domain that has much of its memory
swapped out, and there is insufficient memory on the
source machine to swap the migrating domain's memory
back in from disk.

Seems like a lot of corner cases, and I don't think
these are things we want enterprise customers to
discover the hard way.

I'd suggest setting no_migrate=1 for 4.0, and turning
it off in xen-unstable immediately following 4.0.

> -----Original Message-----
> From: Grzegorz Milos [mailto:gm281@xxxxxxxxx]
> Sent: Sunday, December 20, 2009 8:04 AM
> To: Dan Magenheimer
> Cc: xen-devel@xxxxxxxxxxxxxxxxxxx; Patrick Colp; Andrew Peace; Keir
> Fraser
> Subject: Re: [Xen-devel] [PATCH] Paging and memory sharing for HVM
> guests
> 
> 
> Live migration is something we've got on our to-check lists. There is
> no reason why it shouldn't work properly (albeit slow if many pages
> have been swapped out). We'll test and report back :). In the meantime
> I vote against setting "no_migrate" on, if only because we'll get more
> bug reports , even if it works for us.
> 
> Thanks
> Gregor
> 
> On Fri, Dec 18, 2009 at 6:05 PM, Dan Magenheimer
> <dan.magenheimer@xxxxxxxxxx> wrote:
> >> Does live migration work (reliably) when pages have been
> >> swapped out?
> >
> > Never got an answer on this.  If the answer is NO or
> > not yet (and not likely in time for 4.0), I'd suggest
> > that a domain that enables page sharing should also
> > automagically turn on the "no_migrate" flag.
> >
> >> -----Original Message-----
> >> From: Dan Magenheimer
> >> Sent: Thursday, December 17, 2009 9:20 AM
> >> To: Grzegorz Milos; xen-devel@xxxxxxxxxxxxxxxxxxx; Patrick
> >> Colp; Andrew
> >> Peace; Keir Fraser
> >> Subject: RE: [Xen-devel] [PATCH] Paging and memory sharing for HVM
> >> guests
> >>
> >>
> >> Does live migration work (reliably) when pages have been
> >> swapped out?  IIRC, this is a limitation of VMware's
> >> implementation (or at least was a limitation at one
> >> point).
> >>
> >> > -----Original Message-----
> >> > From: Grzegorz Milos [mailto:gm281@xxxxxxxxx]
> >> > Sent: Wednesday, December 16, 2009 4:15 PM
> >> > To: xen-devel@xxxxxxxxxxxxxxxxxxx; Patrick Colp; Andrew 
> Peace; Keir
> >> > Fraser
> >> > Subject: [Xen-devel] [PATCH] Paging and memory sharing for
> >> HVM guests
> >> >
> >> >
> >> > The series of 46 patches attached to this email contain 
> the initial
> >> > implementation of memory paging and sharing for Xen. Patrick Colp
> >> > leads the work on the pager, and I am mostly responsible 
> for memory
> >> > sharing. We would be grateful for any 
> comments/suggestions you might
> >> > have. Individual patches are labeled with comments 
> describing their
> >> > purpose and a sign-off footnote. Of course we are happy 
> to discuss
> >> > them in more detail, as required. Assuming that there 
> are no major
> >> > objections against including them in the mainstream
> >> xen-unstable tree,
> >> > we would like to move future development to that tree.
> >> >
> >> > Thanks
> >> > Patrick & Gregor
> >> >
> >
>

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel