Good morning Bastien, > > Most solutions to the network flow problem seem to require an accurate view > > of flows at each node, which we do not have. > > Interesting, but for the first hop (local channels) we have the exact balance > available for sending, and for next hops we can consider the channels > balanced (with a random perturbation of X%). The combination of that and > retries could provide interesting results (I plan on testing that on > realistic simulations of the network, I can't know for sure if this will work > until then). > > My first implementation of MPP for eclair uses an algorithm similar to > flocking. > I think your last suggestion of using something similar to `permuteroute` can > be interesting to try too. > I'll give that a shot if we're not satisfied with the results of the flocking > implementation. Is it what you plan on doing for MPP in c-lightning?
Original plan was to derive some kind of `getmultiroute` that does a single graph scan and somehow generates multiple paths, but I kept finding problems in my naive implementations. I never got around to searching for existing algorithms of the needed specifications of getting multiple "good" paths. So maybe `permuteroute` is one idea for implementing multipath we can try to use. Regards, ZmnSCPxj _______________________________________________ Lightning-dev mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/lightning-dev
