These are the bad posts, jbreher.
I'll try the homespun solution, install docker and all.
I'd be interested in your Internet bandwidth consumption, should you be wiling to share.
I will post some data, but don't hold your breath. I'll need some free time to install the thing - starting with Docker - and some more to figure out how to measure LN's bandwitdh tax unbundled from bitcoind's base requirements. Got any suggestions?
Now we get to the "bad boy" part.
My (perhaps flawed) understanding is that in the current implementation, all nodes broadcast all channel state changes to all other nodes. This is necessary, of course, as nobody yet knows how to do decentralized permissionless anonymous path route discovery.
That's incorrect. Nobody yet knows how to do decentralized permissionless anonymous
OPTIMAL path route discovery. Heuristic methods with stored forward tables do work reasonably well. Several mesh-like networks work, without a single tear about being suboptimal.
EDIT: not sure, but optimality (or substantial improvements on route length) could be achieved by permissioning just the network-shaping, route discovery functions. Permissions could be earned by showing reliability and revoked for rogue nodes that systematically misreport.
For those playing along at home, you'll no doubt note that (should my understanding be correct), BW consumption scales at O(n^2).
I'm sure it will improve when the routing invention breakthrough occurs.
Of course - the clueless developers! D'oh, what an unforgivable oversight! The system they implemented is doomed. They chose the SAN (Spam All Network) algorithm for both route discovery and node state update. They slept during their networking lessons, or was it calculus? So they failed to notice that bandwidth grows quadratically with the number of nodes. This way, it's almost as bad as if one increased the block size. But who would ever think of
that?
Out of sarcasm: I'll never believe the protocol broadcasts data mindlessly. On TOR transport, link capacity is precious and doesn't get flooded with irrelevant data. I've never studied the matter in detail, but as an informed guess, I'd go for something like neighborhood broadcasting with empirical discovery of TTL (max distance for broadcasting, initially conservatively low). Once you've found a route that works, you keep it until it stops working - or TOR signals it's time to change circuit. Randomized timing etc. are all nice options to obfuscate things a little more and make network analysis harder.