MPLS Backbone over Frame-Relay Issues

Hey guys, I am hoping somebody can help me understand a bit of an issue I am
having. I am actually just starting out learning MPLS and MPLS VPN
technology, and set out to give it a go tonight in the lab. I have 4
routers connected in a frame-relay hub/spoke topology: R2, R4, R5, R6 with
R6 being the hub. Each spoke router only has a PVC to the hub router, R6,
so any spoke to spoke connectivity will have to pass through R6. For my
MPLS VPN experiment, I wanted the R2,R4,R5 frame routers to be PEs.
Additionally, so that I could fully experiment with a MPLS "cloud" I needed
a P router...since R6 is my frame-relay hub, and all traffic through the
MPLS cloud has to go through it, I figured R6 would be a fine P router.

Off of R2 I have 2 VPNs, one on each fastethernet interface. I call them
VPNA and VPNB (how creative hehe). Off of R4, I have a VPNA CE router. Off
of R5, I have a VPNB CE Router. The CE/PE IGP is RIPv2. Inside of my
frame-relay cloud, I am running OSPF.

Next, I setup MP-BGP between the PE routers, using R2 as a RR. I did the
mutual redistribution into MP-BGP / RIP and everything appeared to be
fine...I could see the proper routes in on all my VPN routers! Awesome.

The problem came when I actually tried to ping...<sigh>. After a long night
of troubleshooting I actually did find the issue, but don't know how to
solve it yet. If I try to ping from a VPNA CE router across the MPLS to
reach one on the other side, it fails. The reason it fails is this from
what I can tell: R2/R5/R6 are all learning the vpnv4 routes through MP-BGP
fine, but they see the next hop as the other end's loopback. For example
R2 (2.2.2.2) peers with R4 (4.4.4.4) and R5 (5.5.5.5) ... so when I attempt
to ping a vpnv4 route that lives in VPNA off of R4 on the other side, R2
sees the next hop as 4.4.4.4 ...it then recursively looks up 4.4.4.4 in it's
routing table and finds a next hop of 150.100.100.4 (R4's frame interface)
and NO LABEL IS FOUND. My LDP session is up between R6 and all my spokes.
My loopbacks are in OSPF...it seems when OSPF advertises from one spoke to
the hub, and the hub (also the DR) advertises that route down to the other
spokes, the next hop does not change....so instead of my next hop in the
recursive lookup being the frame hub , it is the spoke. The entire frame
cloud uses only physical interfaces, and everything in the same subnet.

I hope I explained that well enough for somebody to help out. Has anybody
else out there ran into this? Is it possible to have your frame-relay HUB
be a P router and not run BGP at ALL like this?

Regards,

Joe Astorino
CCIE #24347 (R&S)
Sr. Support Engineer - IPexpert, Inc.
URL: http://www.IPexpert.com

Blogs and organic groups at http://www.ccie.net
Received on Mon Jun 15 2009 - 03:54:07 ART