[c-nsp] IOS XR BFD

Oliver Boehmer (oboehmer) oboehmer at cisco.com
Sun Jul 5 03:00:05 EDT 2009


Nick 'tarantul' Novikov <> wrote on Sunday, July 05, 2009 07:50:

> On Sun, Jul 5, 2009 at 12:40 AM, Pavel Lunin<plunin at senetsy.ru> wrote:
>> Nick, folks are telling clever things.
>> 
>> It is not BGP's deal anyway to control reachability. It's an IGP's
>> task, as well as the best path calculating. Just let IGP carry
>> loopback /32 prefixes, then run iBGP on them, not on subifs. iBGP's
>> job is to carry routes regardless of the topology state.
> 
> Ok. Example of physical topology:
> http://pastebin.ca/1484472
> All physical links protected by IS-IS.
> RR* routers can't keep full BGP table and for this reason ASBR*
> announce 0/0 route only. If I configure BGP session between ASBR* and
> use for it lo0 interfaces I will have a loop. Do not you think?

So you are running iBGP between the ASBRs to announce full routing
between the two, but the ASBRs itself only announce 0/0 towards the RR
(and in turn to the RR-clients). right? But I might be missing something
obvious because I don't see how BFD on the ASBR's iBGP session between
each other is possible (even in IOS) as they're not directly adjacent?

I guess the main problem to address is ASBR1 <-> ASBR2 traffic with RRs
in the middle not having full routing table (I infer this from you
mentioning the loop). This asks for tunnels, so why don't you just
enable MPLS on the ASBRs and the RRs (it seems you already have MPLS in
the core), and then the ASBRs can switch traffic between each other via
the LSP (tunnel).

The second issue is convergence (i.e. failure detection). Running
IGP/ISIS on all nodes (with BFD on the links) sounds possible, ASBR1
will see ASBR2's failure using IGP, and can react accordingly
(invalidates all the routes when next-hop tracking kicks in). 
Tearing down an iBGP session because the BGP-next-hop is gone is usually
a bad thing, it might only be for a second or so in case of a link flap,
and tearing down a full-bgp-feed session only to re-enable it a few
seconds later is not really good use of resources..

> In old school IOS a similar construction works great.

can you post the config in IOS? Not sure I got the full picture.. 

	oli


More information about the cisco-nsp mailing list