[c-nsp] iBGP as MPLS labeling protocol

[adamv0025 at netconsultings.com]

Hi Alireza, 

> Alireza Soltanian
> Sent: Wednesday, January 04, 2017 9:45 AM
> 
> Hi Everybody
> 
> I know this seems stupid, but I want to know is it possible to use ibgp as
> labeling protocol inside of an AS. A scenario which I have in my mind is
> establishing L2VPN connection between two PE routers while Network
> between PE routers uses iBGP(ipv4 afi/safi)?
> 
> Is there anybody here who saw such implementation? I searched the
> Internet and everything I found is about Inter-AS.
> 

Answer to this one is easy, it can be done but no one does this because: 
The most important rule to remember is that BGP speaker allocates Label for
a prefix (VPN label for VPN prefix/Transport label for IP(GRT) prefix) ONLY
when it's changing next-hop(NH) to itself. 
Second rule is you need to create end-to-end Label-Switched Path (LSP)
between your PEs, so every hop on the path (including P routers) needs to
allocate and advertise transport label for the edge PE loopback IPs. 
So in order to do that you have to configure BGP on every hop along the LSP
between edge PEs and run iBGP between the hops and on each hop/BGP speaker
you'd have to manually set "next-hop-self" cause iBGP session is not
changing NH.  
And this is the tricky part cause you might run into bugs with next-hop-self
on iBGP session in combination with RFC3107 (no one tests this).
So easier approach would be to set each hop/BGP speaker along the LSP to its
own AS -as eBGP changes next-hop by default and RFC3107 is tested with eBGP.

PS: 
If you have the possibility to run two BGP processes or multi-AS process on
the edge PEs you should be able to run L3VPN session between edge PEs(and
RRS) as iBGP while transport labels will be carried by chain of eBGP
sessions across your backbone. 
(hmm reminds me of draft-lapukhov-bgp-routing-large-dc and how it could be
converted to MPLS ). 


adam

netconsultings.com
::carrier-class solutions for the telecommunications industry::