Unfortunately, QinQ (which is called "vman" in the Extreme Networks world) does not work with LACP. Corroborating to the problem is the fact that the LAG ports originate on the same switch that provides the transport vmans. Since QinQ transports share src/dst MAC address-space with the connected "customer" VLANs, the "direct" approach does not even work for non-LACP LAGs.

I successfully employed encapsulation of the LACP LAG legs with a pair of EdgeMAX Lite routers, configured for L2 GRE Bridging. I was then able to pipe the encapsulated LAG legs back into the Summit X460 Switch where they originally came from, and transport them over the vmans. Unfortunately, the performance achieved by this solution (~350mbps on a single leg) is far from sufficient.

Two kind members of the NANOG mailing list pointed me to ethernet demarcation devices (E-Line being of interest here) from accedian and ciena, which might just do the trick at GigE wire speed:

http://www.accedian.com/en/products/ethernid-family.html
http://www.ciena.com/products/3902/

Silvan also pointed out that Mikrotik's EoIP might provide adequate performance when provided by a pair of CCR1036 36 core CPU routers. However, it looks as if the accedian/ciena devices would offer a much more cost-effective approach.


Mit freundlichen Grüssen
Chris Burri


---

-= Amat Victoria Curam =-