-
Notifications
You must be signed in to change notification settings - Fork 2.2k
Description
I am not sure if that is directly a bug but I thought the bugtracker is still a better place than the mailinglist.
As far as I understand your documentation of the autopilot it opens channels with nodes that already have most channels open following the barabasi albert model of preferential attachment.
While a scale-free network topology seems reasonable to me I doubt that it makes sense to use that graph model and design decisions for the following reasons:
- It creates power nodes with a large indegree. As long as those nodes don't have enough funding the payment channels will be pretty unbalanced. also it creates a risk that the connectivity breaks when a powernode breaks. Especially a chain reaction of this could happen.
- the power nodes will have to do quite a lot of routing. Also maybe just because they habe been there first. Maybe they don't have the computing power to be connected to millions of payment channels. (nor the funding as mentioned above)
- While social network graphs also follow these kind of distributions I think there are more metrics to consider. For example the number of triangles created (which should be an important property for the lighting network because triangles mean that there are always more that 2 routs in case a channel breaks) This point is similar to the fact that in physical networking we also have routing protocols that use other routing metrics than hops (for example latency, bandwith, some measure of reliability,...). I guess the same will hold true for the lightning network.
- Also in web applications nodes with high indegree usually create quite some chalange for the programmers since they produce so much traffic which needs to be handled.
I am pretty confident we need another network topology that is more stable for the autopilot. In fact I am rather positive that the current form of the autopilot will yield quite some problems in the future.
In case you people agree I'd be very willing to learn go and work on some different strategy to automatically build the lightning network.