NeDi Community

Networking => WAN/Routing => Topic started by: rickli on November 15, 2009, 05:00:01 PM

Title: MPLS with flat L2 backbone
Post by: rickli on November 15, 2009, 05:00:01 PM
The usual approach is probably LDP in an MPLS core. One alternative could be to use a flat L2 network and something like spanning-tree in the core instead. Is anyone using such a setup? What's your experience. Ever had issues? Again I'm just curious, as to what's out there...
Title: Re: MPLS with flat L2 backbone
Post by: tristanbob on November 16, 2010, 07:54:32 PM
Is this a network design question?  Are you talking about a WAN or campus network?

We have a typical (Core, Distribution, Access) campus design.  The core routers connect via routed links to distribution routers, both layers run EIGRP.  The access layer connects via layer-2 portchannels to the distribution routers (2 x 3750s stacked).

Here, I wrote this down in 2006:

I have zero MPLS experience, so I can't help in that area.


Title: Re: MPLS with flat L2 backbone
Post by: rufer on December 09, 2010, 09:06:09 AM
We still have have a flat L2 campus design (roughly standard cisco access/distribution/core design without the L3 core). But we want to change, because with redundancy the spanning tree gets both too complicated or is too inefficient. And even with the new spanning tree algorithms (MST, Rapid PVST) it is too unstable. An example: You have to shutdown a backbone link for maintenance, this can affect (for seconds) the whole network. A partial solution is to use all channels, such that Spanning-Tree rarely ever changes. We didn't have any L2 loops as far as I can remember.

On a side-note: Spanning-Tree MST is good within a VTP domain. I wouldn't do it again with devices not in that VTP domain (or not supporting VTP).

So no, it is not a good design. But if the network is really simple, it is something to consider. Keep it simple and stupid!

I was also thinking about a minimalistic MPLS backbone with PE routers only... need to think about it if it makes sense.