[c-nsp] PBB-EVPN

[Tom Hill]

On 25/08/15 11:28, Ivan Walker wrote:
> As far as I have understood there are two real benefits:
> 1) PBB / MAC in MAC reduces the load on BGP as BGP only needs to worry
> about the B-MACs
> 2) Per-flow redundancy and load balancing

As far as I understand it, #1 is correct (as per regular PBB theory). #2
would depend on your hardware, presumably.

> Is it correct that the only way to connect  a CE device to 2 PEs for all
> active per-flow redundancy is via LACP?  Given that LACP operates at the
> physical port level this seems a little limiting.  I am thinking scenarios
> where the PE-CE link(s) may be some kind of l2 service from  third parties
> delivered over a shared NNI, or perhaps where pseudowire attachment
> circuits are used (into the I-Component).

LACP PDUs are layer-2 and can be transported across Ethernet circuits.
Providing your Ethernet provider will accept & L2PT those frame types,
it is possible to run LACP across the sort of circuit you're describing.
Talk to your providers! If you coupled that with some form of LACP-based
MLAG/MC-LAG/VPC (not all vendor implementations are) then you might be
able to do PE resilience.

AFAIR, SPB-M goes further to remove the multi-homing issues inherent in
PBB, but I've not done any good reading into PBB-EVPN to see if you can
simply multi-home a given C-MAC address and have it work. If it is
hobbled in the same way that PBB is, then yeah, I can see a world where
LACP is the only way forward. Still, it seems oddly limiting.

-- 
Tom