[jerf]

Funny, I thought the exact opposite about the sequence numbers; 16-bit sequence number is cutting it awfully fine for a "unique" ID in the space of a millisecond. Slow-as-molasses Perl can count to 2^16 in 3ms on my machine. Cutting bits out of the node id by hashing MAC, interface name, and a few other things and giving that to the sequence would have made more sense to me. (And you do that hash once at init time.)

In information theory terms, the node ids are carrying log2(number of nodes) bits of information, whereas the node ids are carrying something more like log2(maximum seq id). The latter is almost certainly larger than the former, and at 16 computer-bits allocated for the sequence number overflow would be a real concern of mine, if not on this year's hardware than in a few more years.