Posted on November 14, 2016 8:11 pm
 |  Asked by Jon Nicholson
 |  2545 views
RESOLVED
0
0
Print Friendly, PDF & Email

We having an issue that we believe is related to receiving duplicate ARP requests.

We’ve got nodes (part of openstack) connected to a pair of 7060 switches using MLAG, these 7060 switches then join a VXLan to connect to other pairs of 7060 switches where other nodes exist.

The behaviour we’re seeing with ARP requests is that the broadcast is being flooded to the VTEP that is on both switches in the MLAG group, this is then being forwarded down both legs to the node, so the node sees the request twice.

This seems to be confusing the OVS running on the host.

Is this correct behaviour? Is there something we can do to eliminate the duplication?

My feeling is that this is a bug in the OVS  as it should handle the duplication gracefully, but if we can prevent the duplication from happening in the first place it possibly prevent other problem down the road.

Any help appreciated.

0
Posted by Aesha Parikh
Answered on November 14, 2016 8:41 pm

Hi Jon,

Just to make sure I understood the issue correctly, I am elaborating on the topology description. Lets say host A is connected to mlag pair A and  host B is connected to mlag pair B. Any ARP requests from VTEP A will be flooded to all VTEPs in the floodlist. In this case, one copy will be sent to VTEP B. VTEP B should receive only one copy as MLAG B should have the same vxlan IP shared between peers. From the routing core, there will be a 2-way ECMP for the MLAG B vxlan tunnel IP. Based on hashing, only one peer will receive the broadcast frame (ARP) and will flood it to node B after decap.

Does this answer your question?

0
Posted by Jon Nicholson
Answered on November 15, 2016 10:11 am

Thanks yes I discovered the document:
https://eos.arista.com/vxlan-with-mlag-configuration-guide/

shortly after posting that which explained where I’d gone wrong!

 

Post your Answer

You must be logged in to post an answer.