This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


Re: [Xen-users] XEN - Broadcom issue: survey

> > For an HVM domain, you probably want to look at the tap device rather than
> > the
> > vif.
> What do you mean exactly?
> I'm having the exact same behaviour described by Ivan in his email (dropped
> packets on the vifxx interface), but my tap device hasn't got anything
> strange.
vifX.Y interfaces are only used to send packets to PV network devices
in the guest.  Pure HVM domains (those without any PV drivers) send
all packets over the relevant tapX interface instead.  Errors observed
on the vif interface are therefore completely irrelevant in this case.
If the tap device has nothing strange then you'll have to look
somewhere else.

A couple of things which might be worth investigating:

-- Do you see the same problems with dom0<->domU networking?  If so,
it would be a good idea to fix that before worrying about problems
with the NIC.  Packets which don't need to leave the host don't touch
the physical hardware.
-- I understand you're seeing connections stall for significant
periods of time, and that this happens across a wide variety of
services, yes?  It would be interesting to know if other connections
to the same VM continue working when this happens.
-- Is there a firewall enabled in the guest?  Turning it off might
help.  The dom0 firewall might also be relevant, although that's less
-- If you discover that only off-box traffic is affected, you could
try playing with the offload settings on the physical NIC using
-- You could try checking whether the problem is related to packet
size, using the -s option to ping.  If it is packet size related,
reducing the MTU using ifconfig may help.

> Another strange thing I just observed is that in the host domain the dropped
> packets are on the VIFxx interface on the TX side, while in xm top I can see
> the same number of dropped packets on the "Net0" interface of my guest,
> *but* on the RX side.
> I assumed that this is normal, but maybe it's important.
I don't use xm top's vif statistics myself, so I'm not sure, but I'd
guess it's trying to provide statistics from the guest's point of
view, whereas ifconfig does it from dom0's.  I doubt this is relevant.


Attachment: signature.asc
Description: Digital signature

Xen-users mailing list