Hello everyone,
Question, what are the true benefits to using the next-hop self feature, doesn't matter what vendor.
Most information I see is just to make sure you have reach-ability for external routes via IBGP, but what if all your IBGP knows the eBGP links?
Is there a added benefit to using next hop self in this situation?
Any feedback is much appreciated, either for the question specifically or whatever else you got 
, L3VPN's or underlying technology that has to have that.
Thanks
On an IX, without next-hop-self peer A leaking peer B's routes they receive to
C will have C send direct to B on the IX (assuming flat layer 3 addressing,
and not multiple little /30s or /96s everywhere or something - do any
exchanges do that?)
This may seem more efficient than sending C's traffic to A to get to B (pumping up
the IX's usage graphs) but B may not have peering agreements with C.
Setting next-hop-self avoids this. An 'advantage' in some views. Not related to
n-h-s in an igp specifically, but an interesting (mis)use case.
/kc
I'd like to add that one major advantage is limiting next-hops, thus
labels in your network. This is not just theoretical concern but there
are plenty of practical networks using practical hardware where you
simply cannot expose all next-hops to every node.
Hi
For the MPLS L3VPN the answer is that the next hop attribute needs to be an address from the default VRF and if the peering is happening in a VRF context, there is no address from the default VRF you could use as next hop other than self.
This can be rather inconvenient as there are advantages to have the real next hop. For example fast rerouting works better when all defective routes can be removed in one go because the next hop was removed by the IGP instantly invalidating the affected routes.
Regards,
Baldur