libvirt users - Apr 2014 - Windows 2008r2 guest tcp retransmit hangs

Hi,
I'm currently investigating a problem with our windows 2008r2 guest on 
centos 6 hosts. The issue is that the windows system sometimes sees a 
SYN packet for a tcp connection but doesn't respond. Three seconds later 
the retransmitted packet arrives and this time windows decides to 
proceed normally with the connection.
This is with the virtio drivers but I have now switched to the e1000 
driver where the problem also occurs but not quite as badly as with 
virtio which leads me to speculate that the driver isn't really the 
problem but only affects whatever the real issue is.
Also this does not happen when the systems primary IP is addressed 
directly but only when an IP on the loopback device is addressed through 
a load-balancer.
I've also provisioned a linux guest on the same host and put that into 
the load balancer as well but there everything is fine.

Has anyone run into this and can provide a pointer in the right direction?

Regards,
   Dennis

On 04/08/2014 06:38 PM, Dennis Jacobfeuerborn wrote:
> Hi,
> I'm currently investigating a problem with our windows 2008r2 guest on
> centos 6 hosts. The issue is that the windows system sometimes sees a
> SYN packet for a tcp connection but doesn't respond. Three seconds
later
> the retransmitted packet arrives and this time windows decides to
> proceed normally with the connection.
> This is with the virtio drivers but I have now switched to the e1000
> driver where the problem also occurs but not quite as badly as with
> virtio which leads me to speculate that the driver isn't really the
> problem but only affects whatever the real issue is.
> Also this does not happen when the systems primary IP is addressed
> directly but only when an IP on the loopback device is addressed through
> a load-balancer.
> I've also provisioned a linux guest on the same host and put that into
> the load balancer as well but there everything is fine.
> 
> Has anyone run into this and can provide a pointer in the right direction?
> 
> Regards,
>   Dennis
> 
Hi,

I observed simular issue but was unable to resolve it until migration of
the entire network to OpenFlow, so it was caused by some quirk in the
bridge topology as I think. You may ask in the kvm/qemu list but I
suppose nobody will give a right answer for Centos version because it
cannot relate to any of generic qemu releases.
> _______________________________________________
> libvirt-users mailing list
> libvirt-users@redhat.com
> https://www.redhat.com/mailman/listinfo/libvirt-users