Re: MPLS TE

[Jack Bates <jbates () brightok net>]

On 11/4/2011 12:00 PM, harbor235 wrote:
> I am also looking at FRR which uses a backup tunnel for fast convergence. I
> did however not think
> about the dynamic nature of the tunnel and the potential for
> reestablishment.
Even with primary/secondary paths, the secondary path will normally not 
get used if the primary can resignal to a different path. 
Implementations can get very vendor specific. Each vendor supports 
different subsets of the necessary protocols. I just had a single vendor 
network, that due to lack of SRLG support in their lower end boxes (and 
lack of admin-group in FRR) required setting facility based FRR with 
many bypasses (which luckily they did support at all levels and the 
manual bypasses did support admin-groups for setup).

The only time I usually use dual LSPs is to support load balancing 
across multiple circuits where vendor support is limited (1 LSP down 
each pipe, IGP balance between them, each LSP has secondary path on 
opposite pipe). The idea of MPLS is that the LSP should NOT be down. A 
path might go down and FRR/secondy paths might come into play, but the 
LSP itself should still be handling traffic. Even in complicated QOS 
setups, you can have primary, and multiple secondaries to allow stepdown 
of what a circuit should be reserving, and priorities to even preempt 
circuits to lower class of service (imagine a secondary trying for what 
it currently has without preemption, but then that failing, sets 
different requirements and preempts lesser circuits).

In simpler topologies where I don't need TE and I just want FRR, I've 
been playing with Juniper's LFA implementation. One of my current plans 
is using RSVP/FRR for mpls only services and using LFA for global 
routing. It works for my specific layout, and the only annoyance is 
setting BGP next-hops correctly.


Jack