[Lightning-dev] Testing a Flare-like routing implementation on 2500 AWS nodes

Pierre pm+lists at acinq.fr
Wed Sep 21 14:55:17 UTC 2016


> Excellent work!!
Thanks!

> In reality we envisioned that a DHT would acts as a substitute for a pure
> broadcast network, rather than to allow individual nodes to communicate with
> each other. At the time of the writing of the paper, we envisioned that
> information such as the current onion key for each node, identity keys,
> channel proofs, etc would potentially be stored within a DHT.
Got it, that makes sense. The reference to ip adress was confusing I guess.

> For communications, we referenced that something akin to HORNET could be
> used to allow nodes to communicate with each other [...]
> HORNET was especially attractive as as after its setup phase, there exists a
> bi-directional communication link between two nodes.
Yes I remember you mentioned HORNET on this ml some time ago. I am not too
familiar with it, but I think it operates at the network layer, right? I am
curious how you envision the relation between this layer and the application
layer: would an application-level link (LN channel) be *equal* to a
network-level
link (whatever link there is between two adjacent HORNET nodes if there is such
a thing)? Or would that be completely orthogonal, meaning if A and B
have a channel
open an htlc going from A to B could have to be routed through other nodes?
Does this question make sense?

> Additionally with
> HORNET, the backwards route can be distinct from the forwards route.
That is an amazing feature, kind of magical really!

> As a substitute for the first
> use-case (request/response) the latest design of our Sphinx construction
> could be used, with the requesting node providing the backwards path within
> the e2e payload. It's worth noting that this substitute reveals the entire
> path to the responding node
Yes, that's what we did, it does leak information but we are not sure how bad
that really is.


> Was the capacity of the channels created uniform? Or were the channel values
> uniformly distributed? Or perhaps drawn from a particular distribution?
> Similar to the question above, how were the satoshis requested within the<
> "payment request" distributed?
Well it didn't matter in this particular test, because as you pointed out we
didn't look at channel values (that's what we meant by 'static ranking' and
'undirected graph'). So it could have been anything.
Channel capacities, current values and fees will be key parameters of a future
'dynamic ranking' test, but we didn't think too much about it yet. Maybe normal
distributions? If you have already researched the subject or have any insights
please let us know :-)

Cheers,

Pierre


More information about the Lightning-dev mailing list