[c-nsp] Questions regarding 6PE and route aggregation

[Oliver Boehmer (oboehmer)]

hi.

note: I haven;t touched 6PE in a while, so I might not be 100% accurate:
>
>I’ve been trying to evaluate 6PE as a transition mechanism lately and
>I’ve stumbled upon something I didn’t initially expert. My understanding
>of 6PE is as follows ( and feel free to correct me if I’m wrong ☺ ) :
>PE-A and PE-B peer over IPv4, they exchange routes and labels. Packets
>transit the core over ipv4 and have 2 labels; outer label describes the
>LSP towards the remote PE while the inner label describes the actual IPv6
>next hop ( “CE” or whatever you want to call it ). Label allocation is
>per-prefix ( unless you switch to 6VPE, where you can actually change it,
>with all the consequences that follow ). So, I’ve been thinking, what if
>I don’t want to send the full table to some PEs, but only internal routes
>and a default route ? Internal routes obviously shouldn’t pose an issue,
>but aggregating everything to the default is somewhat trickier. So, I
>fired up my lab, got an ebgp session with an upstream and…here’s what
>happened:
>
>Scenario A, get a default route from an upstream and advertise it
>downstream while suppressing all other external routes: Everything works
>fine. Would it work fine with >1 upstream though, or would I get
>suboptimal routing, as the PE with the full table would just perform a
>lookup on the inner label ( which would correlate to whatever single
>default route was preferred from upstreams ) ?

yes. you can check this by looking at the LFIB entry of the label BGP
assigned to the ::/0, it should point straight out to the egress interface.

>
>Scenario B, no default route from upstreams. Just generating one towards
>downstreams with default-information originate: Control plane looks ok ,
>data plane just fails!

how does the RIB/FIB/LFIB look like? How exactly do you generate the
default? via "neighbor .... default-information-originate"?

>
>Scenario C, same as scenario B apart from the addition of a static
>default to Null0 on the PE originating the default route: Surprisingly,
>it works. Is it supposed to work though ? If so, should I assume it’s
>doing 2 lookups ( one of the label and one for the actual prefix ) with
>all the performance implication this carries ?

My guess is that the PE originates an aggregate label, which tells the PE
to do another lookup. This is just a guess..

>
>Also, any advise on how you tackled similar issues would be greatly
>appreciated. Assuming scenario C is a valid design, it feels kinda
>wasteful. 

why?

BTW: There is a thread on https://supportforums.cisco.com/thread/2006006
about this..

	oli