[c-nsp] DCI interconnects at L2 vs Encapsulated spanned Vlan

[Peter Rathlev]

On Tue, 2016-04-19 at 08:12 +0000, Nick Cutting wrote:
> If you use a L2 tunneling protocol over a L3 DCI - does this mitigate
> all the L2 risks of a data centre interconnect?

Not as such. The tunnelled packets have a TTL header and loops in the
core are thus less of a problem, but L2 loops through the tunnel can
still persist.

If you have for example a set of VSS switches in both ends you can
avoid loops altogether, but that goes for both mechanisms. And not
everyone is equally happy about VSS/stacking/IRF and their ilk.

> i.e. Would using encapsulation of the L2 frames be much better than
> for example, running 3 Vlans over the link, using one for routing and
> 2 for spanned vlans? 

I would personally prefer using the 3 VLANs over a trunk. A tunnelling
mechanism introduces complexity that may outweigh the benefits.

One benefit is the ability to re-route on link down fast enough that
things like STP will not notice. You will see higher latency as long as
the traffic is re-routed but no topology changes in STP.

We use connections of both types, L2 with VLANs and L3 tunnelled via
EoMPLS. We use the L3 tunnelled connections where the distance between
the DCs makes it difficult and/or expensive to have a direct L2
connection. We haven't really seen any problems with these tunnels.
Their ability to survive connectivity problems via re-route has been
very nice for us.

-- 
Peter