NLSR

I have tested this on 34 node, 58 node, 78 node, and 99 node topologies - all of them converge.

There is no difference in convergence time with or without /localhop patch.

For 78 node:
with /localhop LSA: NACKs = 0 duplicate NACKs
w/o /localhop LSA: NACKs = 205359 duplicate NACKs

Then the outstanding question is whether this is a "best practices" improvement. If there is no performance gain, there ought to be some kind of code quality gain or similar. Otherwise this patch should be dropped.

Some more statistics:

For 58 node topology:

1) Without /localhop:
2310994 LSA Interests

464960 NACKs = 442760 Dupl NACKs + 22200 NoRoute NACKs

2) With /localhop
141312 LSA Interests (16 times fewer than without /localhop)

77820 NACK NoRoutes

(Note: this is using Multicast strategy which has NACK processing and sends leastSevereNack downstream when all outRecords have NACKed).

Maybe better idea is to do this:
https://redmine.named-data.net/issues/3626

Reason for so many NoRoute NACKs:
LSA prefix is on multicast strategy and all the neighbors which don't have the data receive an interest.

Is it correct behavior for the other nodes to return a NoRoute if the prefix is /localhop?

Comments from Lan:
My understanding is if you want to forward an interest and there’s no FIB entry, then send back a NoRoute. Here /localhop means don’t forward it further so the precondition for NoRoute doesn’t hold.

NFD call on 01/03/2017:
If we send an interest then we except data back, so NACK NoRoute is semantically correct.

Correct solution is send interest to the face where the sync update was received from:
https://redmine.named-data.net/issues/3626