[ovs-dev] [BUG] broad-/multicast & SLB bonding -> FAIL

Markus Schuster ml at markus.schuster.name
Wed Jan 23 09:33:49 UTC 2013


On Tuesday 22 January 2013 22:26:33 Ben Pfaff wrote:
> On Tue, Jan 22, 2013 at 07:39:01PM +0100, Markus Schuster wrote:
> > Open vSwitch sends out the multicast frames on ALL physical interfaces
> > belonging to the SLB bond. That causes a lot of confusion on the physical
> > switches that XCP hosts are connected to (VM MAC addresses jumping
> > between ports multiple times a second).
> 
> It shouldn't be doing that.  How did you determine that this is
> happening?

First of all a little overview of the components involved, so we're actually 
talking about the same thing:
http://www.picpaste.com/pics/overview-4xoHXl1v.1358932433.png
All XCP hosts are Blades, connected to the BladeCenter internal switches 
(Nortel L2-3 switching modules) which are connected to our switching backbone 
(via LACP aggregated links, but shouldn't matter for this issue). 

How did we determine this is happening: First we had a look in the switching 
table on Backbone Switch 1, seeing that the MAC address of the VM in question 
running on the XCP Blade jumps between the uplink ports to BC switch 1 and BC 
switch 2 like crazy. 
Next, we configured port mirroring on BC switches 1 & 2 to see all traffic of 
the BladeCenter internal port that connects the XCP Blade to the BC internal 
switches. We dumped traffic with tcpdump and had a look with Wireshark 
afterwards. What we saw was the exact same frames carrying the IP multicast 
packets on both BC switches at the very same time. 
Next, we connected directly to the XCP Blade, started tcpdump on the two 
physical interfaces forming the SLB bond and filtering for broad-/multicast 
traffic and the MAC of the VM in question. Again, we saw broad- and multicast 
traffic going out on both interfaces. 

That's the story, hope that helps you somehow. 

Regards,
Markus



More information about the dev mailing list