NLSR

I have developed a plan for implementation: I intend to give the NPT a storage pool of routing table entries in the form of a hash map. I have not yet decided whether to store the RTEs in the hash set as pointers or as whole objects, but I am not sure that this distinction is very important.

Keeping an array of pointers would be more space efficient, since each pointer is only 8 bytes, but because hash maps have very fast lookup time and also this approach solves the entry duplication problem, I think it is worth the slight loss of space efficiency.

I intend to use the names associated with the router to hash the RTE pool entry, which contains the next hop list, count of NPT entries using it, and all the other things a RTE would have, to map the values. Again, I think the ease of use easily trumps the small costs.

After some time I was able to iron out kinks with the convergence experiment I'd developed. A reduced testbed topology of 30 nodes was used. The working directory was made on a ramdisk, and before each experiment iteration the whole ramdisk was emptied, stale mini-net furniture was destroyed, and any lingering NFD processes were killed. The results were:

Master NLSR. 30-node topology, on Fri Aug 26 14:02:35 EDT 2016: average time over 10 runs: 28.73963184357 seconds.
NPT-patched NLSR. 30-node topology, on Fri Aug 26 14:15:51 EDT 2016: average time over 10 runs: 28.52513318061 seconds.
Difference: ~0.214s.

With a difference this small it is not possible to tell if this is statistical noise or a significant result. The size and/or complexity of the topology could be increased, or more iterations could be done to smooth out possible statistical variations.