[pve-devel] Windows live migration and time problems

[Alexandre DERUMIER]

>>Maybe because we use the hyperv enlightment hv_time for win >= vista ? 
>>as far I known, it's do almost the same than kvm_clock for linux guests. 

Also, hv_time works only with tsc clock, so I think this is why libvirt disable hpet


----- Mail original -----
De: "Alexandre Derumier" <aderumier at odiso.com>
À: "pve-devel" <pve-devel at pve.proxmox.com>
Envoyé: Vendredi 9 Septembre 2016 14:37:52
Objet: Re: [pve-devel] Windows live migration and time problems

>>As for the time speed issue mentioned in [4], so far we've only 
>>encountered this with windows XP. 

Maybe because we use the hyperv enlightment hv_time for win >= vista ? 

as far I known, it's do almost the same than kvm_clock for linux guests. 


----- Mail original ----- 
De: "Wolfgang Bumiller" <w.bumiller at proxmox.com> 
À: "pve-devel" <pve-devel at pve.proxmox.com> 
Envoyé: Vendredi 9 Septembre 2016 14:18:46 
Objet: [pve-devel] Windows live migration and time problems 

Several users[1][2] have run into live migration issues with windows 
machines which seems to have to do with windows' time handling[3][4]. 

According to a reply to [3] it seems to be possible to "fix" this by 
enabling HPET (iow. removing the `-no-hpet` option) for windows 
machines, which on the other hand might cause issues described in [4]. 

We've been able to reproduce this on both PVE and Ubuntu with libvirt. 
In both cases we used a windows 10 machine with a virtio-scsi disk, 6 
cores, 2G RAM and virtio-network. Changing any of these components 
influences the probability of these hangs which makes it incredibly 
hard to find a real solution. 
The issue manifests in a hanging graphics stack. The machine still 
responds to pings - if you have a command line open you might even be 
able to type some commands there (ie. a `ping` to the outside works - 
partially). 

As for the time speed issue mentioned in [4], so far we've only 
encountered this with windows XP. Curiously we start adding `-no-hpet` 
with win vista/7 and up (same as libvirt does AFAIK, at least in my 
tests I did see the -no-hpet option). 
This makes me wonder if we should move this option to win-xp or 
remove it altogether hoping that newer / more up to date windows 
versions are less fragile in their time management? (Note that XP is 
old, outdated and unsupported, so I don't care much about what happens 
to it on a migration or how fast its clock is running... the setting 
could be made optional in any case.) 

It would be nice if some more people could test this and maybe give 
some feedback (and possibly share their findings here on the qemu bug 
tracker in [3] and/or [4]). 

[1] https://forum.proxmox.com/threads/windows-vm-hang-after-live-migration.28216/#post-145128 
[2] https://forum.proxmox.com/threads/live-migration-hungs-windows-guest.29101/ 
[3] https://bugs.launchpad.net/qemu/+bug/1618431 
[4] https://bugs.launchpad.net/qemu/+bug/599958 

_______________________________________________ 
pve-devel mailing list 
pve-devel at pve.proxmox.com 
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel 

_______________________________________________ 
pve-devel mailing list 
pve-devel at pve.proxmox.com 
http://pve.proxmox.com/cgi-bin/mailman/listinfo/pve-devel