[j-nsp] Proper Break of MPLS RSVP Ring

[Chris Kawchuk]

Post relevant configs and an actual diagram (Visio -> PDF)

Without this, anything we say is pure speculation -- and we end up playing '20 questions' with you. Getting an MPLS/RSVP/LDP/IGP/BGP/Mesh/TE network setup involves multiple steps and config-knobs being turned on and turned on correctly. Missing any one of them can result in undesirable behaviour.

1. RSVP priority/preference (?) has no bearing on forming an MPLS forwarding path adjacent between two LSRs.

2. There is no "break" in an MPLS network if it happens to be attached in a ring. This is not Spanning Tree. You have a fully routed network. Topology can be arbitrary.

3. How are you setting "broken RSVP down?" RSVP only goes "up" to a neighbour IF it actually has a reason to talk to it's neighbour. If you do not book an RSVP LSP across the link (due to ERO or following the IGP to the egress point), the two LSR's never exchange RSVP packets, because they have no reason to do so. This is known and expected behaviour. This is not LDP, which is 'chatty' and tries to reach out and touch it's neighbour and dynamically create FECs and transport label tables. RSVP only is invoked on an LSR-LSR link if an actual reservation needs to be made on that link.

4. What does your IGP suggest about the shortest path in the topology?

5. do you have family mpls enabled on all the relevant interfaces?

6. do you have all the relevant interfaces you want to run rsvp on, declared in protocols rsvp, and protocols mpls?

etc.. ;)

- Ck.




On 22/07/2015, at 5:18 AM, Levi Pederson <levipederson at mankatonetworks.net> wrote:

> All,
> 
> Double Checked the Layer 2 ring today and it seems solid.
> 
> Once again we have B and C co-located and A and D in remote locations with
> a link between them.
> 
> Currently there is no RSVP between C and D and this is making my ring go
> right instead of left!
> 
> I can Ping from D to C (it's next hop on the ring) if I force it out the
> MPLS interface.  However when I ping the LSP interfaces (loopbacks) it
> takes the long way around).  Short is 10ms and the long goes up to almost
> 26ms (pinging loops , again the long way around).  Current production
> traffic backs this up.
> 
> This leads me to believe there is not a Layer 2 issue but something more
> enigmatic.
> 
> Currently reading up on RSVP priority/preference but that seems like taking
> a 2Ton Electromagnetic Sentient WreckingBall to hammer in a nail.
> 
> Thank you,