So in the following case:

- source attached to V-hub2
- receiver attached to V-spoke1
- V-spoke1 associated with V-hub1

I was questioning whether it is really necessary to have a data path of
source-->V-hub2-->V-hub1-->V-spoke1-->receiver, or whether we could have the
more optimal data path of source-->V-hub2-->V-spoke1-->receiver

If I understand correctly, the answer is that in order to obtain the more
optimal path:

1. We would need a new procedure for determining which packets need to be
   discarded because they come from the "wrong place"; this would have to be
   tunnel-based rather than sender-based.  (A similar issue arises in
   extranet.) 

2. This procedure would require more BGP routes to be sent from a v-hub to
   its associated spokes, since in hub has to tell its associated spokes the
   tunnel to use for each (S,G).  (This makes it difficult to use I-PMSI or
   (*,*) S-PMSI A-D routes in the hub-to-associated-spoke control plane.)

While I agree that these points are valid, I don't see 1 as a real issue,
and 2 is a common tradeoff (increasing the optimality of the data plane
requires more control messages).  The issue of unsolicited flooded data is
really a sub-issue of 2.

However, in the context of WG LC, I think you have adequately answered the
question of why it makes the choice it does.










  

  

  

  

  

Reply via email to