[c-nsp] VRF-Lite Question

[Phil Mayers]

Jeff Kell wrote:
> After some experimentation, I'm more confused than ever.  So to
> generalize this even a bit further:
> 
> Jeff Kell wrote:
>> Siva Valliappan wrote:
>>   
>>> Yep.  GRE tunnelling is not supported on the C3550, C3560, C3750,
>>> C3560-E, C3750-E.  So please do not use it.
>>>     
>> Which begs the question, *is* there an officially supported method of transport between a 3550/3560/3750 CE and a non-directly connected PE?  
>>
>> If directly connected there is always the dedicated vlans of a trunk (or subinterface) solution, but is there any other plain old layer-3 solution?
> 
> Well, even if directly connected you can't use subinterfaces.  From
> http://www.ciscotaccc.com/lanswitching/showcase?case=K30523495 "Do not
> configure Ethernet sub-interfaces on a switch. Configure it as a trunk,
> and terminate the L3 IP interface on a Switched Virtual Interface (SVI)."
> 
> That's fine for single point-to-point hub-and-spoke, but gets downright
> ugly if you're trying to setup an L3 mesh.

Well, it's tedious typing but hopefully you're only doing it once and 
you're running all your configs through some kind of validation 
procedure (or using a deployment tool) right? Oh wait I remember, this 
is IP networking and we're still stuck in the 1970s for tools.

> 
> OK, here's a real-life example... a stack of 4 3750s with 4 VRFs spread
> across them (OOB network management/monitoring, system admin, back-end
> databases, and public/DMZ) to uplink to redundant 6509 PEs.  What's the

Lots of per-VRF L3 point-to-points, routing processes and routing 
adjacencies is the only working method on non-MPLS capable routers.

It's tedious, but then the 3550/3750s are cheap.

We are looking at J63xx+L2 switch for new locations.

> optimal uplink configuration for performance and redundancy? 
> Etherchannel?  Redundant L3?  What if it's two stacks of two 3750s?

Depends on your needs. We use etherchannel because it detects failures 
and "reconverges" faster, but of course that has failure modes which 2x 
point-to-point does not. Also, you obviously can't etherchannel to 2 
separate 6500s.