[c-nsp] IOS XR BFD

[Nick 'tarantul' Novikov]

On Sun, Jul 5, 2009 at 4:59 PM, Pavel Lunin<plunin at senetsy.ru> wrote:
> You know, I might also be missing something, but I don't see much difference
> from iBGP's point of view. Traffic anyway goes through RRs on the way from
> the core to outside as well as between ASBRs and RRs know only defaults.
> What advantage does iBGP on subifs give here? I'd understand if you had a
> link or an LSP between ASBRs and wanted to exclude a possibility of passing
> plain IP traffic from one ASBR to another through RRs, but in this case...
> am I missing something? How loops are avoided now?

1. RR1 have 0/0 from ASBR1 (as best route) and send packets to RR1
2. ASBR1 have BGP session with ASBR2 and destination prefix close through ASBR2
3. ASBR1 send packets back to RR1
4. Go to p.1

Ok, I can configure separated L2 path for ASBR1-ASBR2:
ASBR1 -trunk- RR1 - xconnect - RR2 -trunk- ASBR2
But if xconnect fails, I get the situation described above for the BGP
timeout (3*60 seconds by default)
To avoid this possible to configure the BGP session from ASBR subif
and use BDF for fast session drop if L2 connect fails.



> Moreover what is a reason of separation of RRs and ASBRs in such a manner?

It is easier to operate.

> Normally you want RRs to carry traffic as little as possible but do well
> their control plane jobs with no excuse. Why RRs can't fit full BGP? I bet
> because their FIBs are constrained (sort of sup32 TCAM capability problem),
> but not due to thier RIBs. Ideally RRs should stand out of forwarding
> topology and not carry transit traffic at all.

RR is a 7600 with notXL RSP. 256k prefixes only.

>> otherwise static routes will save you (does
>> > IOS XR support BFD for them? :)
>>
>> So fsck... No. IOS XR can't. If I configure (X.X.X.X - subif BGP
>> neighbor, not lo0 address)
>> router static
>>  address-family ipv4 unicast
>>  X.X.X.X/32 Null0
>>  !
>> !
>> BGP session don't drop!
>> In old school IOS a similar construction works great.
>
> Hm...  seems strange anyway. Does this route come active? Isn't it possible
> that something like an ARP entry for x.x.x.x treated as a connected route
> with lower admin distance? I know some non-cisco devices which can do so. Or
> something else might beat this static route. What about 'sh ip route
> x.x.x.x' (or whatever this command looks like in IOS XR) and 'ping x.x.x.x'
> after adding this route? And if the route to null0 comes active and ping
> fails, but iBGP stills alive, can you do some sort of investigation to know
> how traffic reaches the peer, which path it goes along?

sh ip route indicates Null0, ping work. Static route /32 longest match
than connected to interface /30. I think the traffic should go to
Null0 and ping must be broken (and BGP session).
However, this design does not help me. Oldschool IOS for my ASBR (12k)
don't support BFD for static route feature (but IOS XR support it).

And my question is not how I should be in this situation.
What is the logical explanation that BFD does not work in internal neighbors?

-- 
tarantul
Dios es Amor