Statistics on the latency between nodes for different network sizes


#21

Estimate about how many lines of code it is now or how many K? Hoping its still the size of DOS 1.0


#22

No clue sorry and its not always a good indication. Bugs destroy that


#23

As in fixing a bug can balloon things?


#24

Care to weigh in again?


#25

This was a really great post, vaguely remember seeing it but nice to read again. Nice necro!

I wonder what the results would look like with adding ‘server centers of population’ … server instances could be hosts and centers would look more or less like additional cities. I guess unless the server centers were a high proportion of the network, the chance of a ‘rack to rack’ hop would be quite small… the results for very large networks would probably be similar, but I bet that the initial distribution will be quite server heavy and the early network will have lower latencies as a result.

Server-“server in a different center” hops will also be close to the ‘direct connection’ model, so I would imagine the more accurate result will be a lower latency early network, x1.45 on everything, and as the network grows, a constant /hop malus will phase in (router latency) and some multiplier for the increasing probability of non straight line connections. Still though, these results are quite hopeful, it is better than I would have guessed. A good caching implementation will be key to make the network snappy, and nobody will mind waiting a few seconds to connect to their backed up photos.

How does the number of hops affect data transfer (e.g. large files?) is there some optimization of the routing post-contact? Or is a large number of high latency torrent sources a better mental model?