MST and QinQ

Unanswered Question
Jul 23rd, 2007

In a test lab, I have implemented a backbone using 4 6500 switches connected in a square configuration. The backbone is using MST, the four 6500 are in the same MST region. Then, i connected two 2950 switches to the backpone. The 2950 are not using MST, they are using PVSTP.

If I simulate a failure in one of the backbone links, the connectivity between the two 2950 converges in one second and all is OK.

If I define QinQ to transport a 2950 VLAN thru the MST backbone and I simulate the same backbone kink failure, connectivty between the two 2950 is lost for more than 30 seconds. The interface between 6500 amd 2950 is re-initialized every time the backbone topology changes.Is there any known incomptibilty between MST and QinQ?

Moreno sartorel

I have this problem too.
0 votes
  • 1
  • 2
  • 3
  • 4
  • 5
Overall Rating: 0 (0 ratings)
Francois Tallet Mon, 07/23/2007 - 08:32

If you are sending traffic in vlan X from one 2950 to the other, without QinQ, this traffic is indeed carried by vlan X on the backbone. Now, if you are using QinQ, all the 2950 traffic is mapped into a vlan Y before being transported in the backbone. So maybe the failure you were doing in the backbone was impacting vlan X and vlan Y differently? That could explain the difference.

The interaction between MST and PVST are done using legacy STP rules. That means that if MST has to sync the ports leading to the 2950 (which is quite likely), they will block for 30 seconds. Traffic will be interrupted, regardless of whether you are doing QinQ or not.

If you are running QinQ and l2pt, you could put the MST ports leading to the 2950 in portfast mode. This is because there is no interaction between the STP run on the 2950 and the MST region in that case. The 2950 are responsible for detecting the loop they introduce by sending PVST BPDUs (tunneled) across the core.



msartorel Mon, 07/23/2007 - 08:52

As i tried to explain, the network behaves in different way depending if we are using QinQ or not. Using the same topology, without QinQ, traffic on vlan X between the two 2950 is interrupted only for 1 sec, if we mapped lan X on lan Y, traffic is interruped for about 30 sec.

Moreno sartorel

Francois Tallet Mon, 07/23/2007 - 09:00

Please, provide more details. What is the STP state of the ports involved in the forwarding of the traffic between the 2950 during the problem. Provide information for both vlan X and Y so that we can see if there is a difference.

QinQ mainly change the data path, there is no MST code related to QinQ. When doing QinQ, you are not running STP with the edge bridges. Considering that the interaction with PVST is roughly the worst case scenario for MST, that cannot be bad;-)



msartorel Tue, 07/24/2007 - 04:52

We will try using RSTP on the 2950 and we will let you know the results. Thanks.

Francois Tallet Tue, 07/24/2007 - 08:42

I'm afraid Rapid-PVST will not change much. When interacting between Rapid-PVST and MST, we still fall back to the PVST model. It would only help if you were not trunking on the link between the 2950 and the MST switch. And anyway, this interaction is only relevant to the case where you are not running QinQ. When you have QinQ configured, there is no interaction between the STP on the 2950 and the core MST switches. That's why I recommended using portfast on the MST switches.

Let me know the outcome.




This Discussion