[c-nsp] BGP Next-hope convergance

[Mikael Abrahamsson]

On Mon, 8 Sep 2008, Peter Rathlev wrote:

> Does this have an effect on the convergence times? I seem to remember 
> having been told that this is a good idea generally, but never really 
> understood why. Can anybody shed light on why this is?

What Ytti said, and let me give you an example:

PE1 and PE2 uplinks a redundantly connected customer.
PE3 has another customer connection in the same VPN.

Using unique RDs per PE, will ensure that PE3 has routes to both PE1 and 
PE2 in a route reflector structure, and if PE1 goes away then PE3 will 
have PE2 route in RIB and can update FIB without RR involvement.

In case PE1 customer link goes down, it can change FIB to point to PE2 
without RR involvement meaning packets will be forwarded continously as 
soon as the link-down is detected and FIB is updated, instead of the RR 
noticing it and sending updates.

So yes, you want to use PE unique RDs, there little downside apart from a 
bit higher memory usage.

-- 
Mikael Abrahamsson email: swmike at swm.pp.se