cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Announcements
Join Customer Connection to register!
637
Views
0
Helpful
1
Replies
msundman78
Beginner

dot1q-tunnel mac table convergence delay

I've configured two switches (A and B) todo dot1q-tunneling (Q-in-Q) for the normal dot1Q trunk between switch C and D. In addition I have a direct link between C and D. Everything works as expected and at normal operation Gi0/24 on switch D becomes blocking by RSTP, and traffic flowing fine between my PingPC 1 and 2.

If I then disconnect the direct link between C and D, RSTP converges instantly and Gi0/24 on both C and D goes into forwarding state.

The problem however is that during normal operation when all links are up, switch A learns PC2's mac address on Gi0/24, and when the direct link between C and D goes down switch A don't know anything about that. None of its interfaces has flapped and he doesn't participate in C and D's STP conversations, so A maintains its MAC table which sais that PC2 is still available on Gi0/24.

So when I try to ping PC2 from PC1 the packets arrive at Sw A but are just bounced back out on Gi0/24, until like 20-30 seconds later when PC2 has broadcasted something or the MAC table on Sw A has expired so it re-learns that PC2 is now available on Te1/2 and traffic is resumed.

If I manually clear the mac-table on Sw A, it obviously starts working instantly.

So, what can I do to improve this situation to get as fast convergence as possible?

dot1q-tunnel.png

1 REPLY 1
msundman78
Beginner

I just found out that I could disable mac address learning completely on the Q-in-Q switches (A and B) with

no mac address-table learning vlan XXX

which seems to solve the problem. Now I only drop packets for 1 sec before traffic is resumed again. Wonderful

Is there any draw-backs of doing that in the topology descibed above? I basically just want the A and B switches to "light" a fiber and transparently pass all traffic between switch C and D, so it would seem okay to me to disable mac learning in such a topology.

Does it cause any more CPU overhead or something to forward frames not using a mac table?

Googling around, I can't seem to find anyone else recommending disabling mac learning which makes me a bit worried. I can't be the first one having this problem...