[Qemu-devel] SLIRP migration?

[Qemu-devel] SLIRP migration?

[Dr. David Alan Gilbert]

Hi Samuel, Jan,
  I'd like to convert SLIRPs migration code to VMState
and avoid all the qemu_put_/qemu_get_'s - but first I'd like to understand
when a lot of the slirp migration code makes sense.

I can see that migrating a VM with slirp makes sense (actually more
commonly saving a VM with slirp to a file); but when that happens
aren't you likely to lose all the network connections anyway?
If that's the case then why bother saving all the TCP state?

Under what circumstances do the network connections survive and
how can I test that?

Dave

--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK

Re: [Qemu-devel] SLIRP migration?

[Jan Kiszka]

On 2016-09-28 20:00, Dr. David Alan Gilbert wrote:
> Hi Samuel, Jan,
>   I'd like to convert SLIRPs migration code to VMState
> and avoid all the qemu_put_/qemu_get_'s - but first I'd like to understand
> when a lot of the slirp migration code makes sense.
> 
> I can see that migrating a VM with slirp makes sense (actually more
> commonly saving a VM with slirp to a file); but when that happens
> aren't you likely to lose all the network connections anyway?
> If that's the case then why bother saving all the TCP state?
> 
> Under what circumstances do the network connections survive and
> how can I test that?

I think I never tried this myself back then, nor do I remember any
success story about slirp live migration from the top of my head.

Jan

-- 
Siemens AG, Corporate Technology, CT RDA ITP SES-DE
Corporate Competence Center Embedded Linux

Re: [Qemu-devel] SLIRP migration?

[Dr. David Alan Gilbert]

* Jan Kiszka (jan.kiszka@siemens.com) wrote:
> On 2016-09-28 20:00, Dr. David Alan Gilbert wrote:
> > Hi Samuel, Jan,
> >   I'd like to convert SLIRPs migration code to VMState
> > and avoid all the qemu_put_/qemu_get_'s - but first I'd like to understand
> > when a lot of the slirp migration code makes sense.
> > 
> > I can see that migrating a VM with slirp makes sense (actually more
> > commonly saving a VM with slirp to a file); but when that happens
> > aren't you likely to lose all the network connections anyway?
> > If that's the case then why bother saving all the TCP state?
> > 
> > Under what circumstances do the network connections survive and
> > how can I test that?
> 
> I think I never tried this myself back then, nor do I remember any
> success story about slirp live migration from the top of my head.

It certainly survives - i.e. if you migrate a VM with slirp
the VM carries on running, and then you can make a new connection
to the VM and it works.

Maybe the issue here is that you need to store a lot of state
so that the guests view of the dieing connections after migration
is correct?

Dave

> Jan
> 
> -- 
> Siemens AG, Corporate Technology, CT RDA ITP SES-DE
> Corporate Competence Center Embedded Linux
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK

Re: [Qemu-devel] SLIRP migration?

[Samuel Thibault]

Hello,

Dr. David Alan Gilbert, on Wed 28 Sep 2016 19:00:45 +0100, wrote:
> I can see that migrating a VM with slirp makes sense (actually more
> commonly saving a VM with slirp to a file); but when that happens
> aren't you likely to lose all the network connections anyway?

Yes.

> If that's the case then why bother saving all the TCP state?

Because there are not only network connections, actually the network
connections are *not* saved: slirp_state_save iterates only over
exec_list with ex_pty = 3, i.e. guestfwd to a device.

Samuel

Re: [Qemu-devel] SLIRP migration?

[Dr. David Alan Gilbert]

* Samuel Thibault (samuel.thibault@gnu.org) wrote:
> Hello,
> 
> Dr. David Alan Gilbert, on Wed 28 Sep 2016 19:00:45 +0100, wrote:
> > I can see that migrating a VM with slirp makes sense (actually more
> > commonly saving a VM with slirp to a file); but when that happens
> > aren't you likely to lose all the network connections anyway?
> 
> Yes.
> 
> > If that's the case then why bother saving all the TCP state?
> 
> Because there are not only network connections, actually the network
> connections are *not* saved: slirp_state_save iterates only over
> exec_list with ex_pty = 3, i.e. guestfwd to a device.

Ah! Thanks for the explanation, that makes a lot more sense.

Dave

> 
> Samuel
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK