[f-nsp] SuperX with RSTP causing some loss on topology change

Nicolaj Kamensek nsp at accelerated.de
Wed Dec 12 19:07:52 EST 2012


Hello list,

after having installed a new setup in a network, I noticed a more or 
less big issue for which I wasn't able to spot the cause yet. This is 
the setup:

+------------+       +------------+
| Juniper MX |-------| Juniper MX |
+------------+       +------------+
       |                    |
       |                    |
       |                    |
+------------+       +------------+
| Brocade SX |-------| Brocade SX |
+------------+       +------------+
   | | . . . |
   T T . . . T
   o o . . . o
   R R . . . R ...


The Juniper MX routers are acting as Layer3 switches with RSTP enabled, 
one being the root bridge, the other one being the backup root.
The SX devices are both SX400 with RSTP enabled as well and a priority 
of 16k. The ToR switches are connected to the SX devices, usually with 
stp root-protect configured on each downlink interface. What happens is, 
that when a new port is connected or relocated between the two SX 
switches, I see some major packet-loss for a few seconds on all the 
ports behind the SX in question.

I am not sure what causes this. One assumption was a topology change 
resulting in a flush of the mac table but since I can't seem to be able 
to find out if or if not a TC occured (what is the command on the SX if 
available??), I am not 100% sure that this is related to the TC.
Another colleque's opinion is that the rstp implementation on brocade 
devices is very poor and recommends to migrato to mstp which works much 
better. Although I am willing to test the mstp proposal, I'd really like 
to here some other ideas or maybe definitive answers to that behaviour 
and ideas how to solve this.

Disabling STP on the downlinks is not an option. Would declaring the 
downlink-ports to admin-edge-ports help in this situation? Since some 
ToR switches have (R)STP enabled theirselves, wouldn't the arrival of an 
BPDU package disable that configuration anyway?

Just to have it mentioned, this is the relevant configuration snippet on 
the SX devices:

spanning-tree single 802-1w
spanning-tree single 802-1w priority 16384

So I am actually using the correct RSTP protocol (which is not the case 
with spanning-tree rstp).


Thanks for any hints!


Best regards,
Nico




More information about the foundry-nsp mailing list