On 5/26/22 23:32, René Pickhardt via Lightning-dev wrote:
Dear fellow lightning developers,
please note my recent blog article titled "Price of Anarchy from
selfish routing strategies on the Lightning Network" [1] where we
investigate how the selfish behavior of nodes sending Bitcoin over the
Lightning Network may lead to higher drain on channels which in turn
is expected to result in higher depletion and failure rates for
payments on the network. All of the observations have been derived
purely be looking at statistical measures and computations on the data
that the Gossip Protocol and Bitcoin Network provides about the
topology of the Lightning Network. No probing or empirical experiments
had to be conducted to derive these theoretical results. All code can
be found in the lnresearch repository at [2].
...
I hope the described effects won't be too strong for the expected
traffic and usage of the network so that the technology will work
properly at the required scale. I am very happy for your thoughts,
feedback, comments and questions as I find it fascinating to see how
the game theory of the Lightning network will eventually play out and
at least in my current understanding seems to produce limitations to
the amount of traffic the protocol may eventually be able to handle.
with kind Regards Rene Pickhardt
[1]:
https://blog.bitmex.com/price-of-anarchy-from-selfish-routing-strategies/
[2]: https://github.com/lnresearch/Price-Of-Anarchy-in-Selfish-Routing
Dear René,
a few years ago I made a very small contribution to this list by posting
a paper on "Modeling a Steady-State Lightning Network Economy":
https://lists.linuxfoundation.org/pipermail/lightning-dev/2019-August/002115.html
I mention it here because perhaps there are some ideas tangentially
related to the research program you are conducting on routing
strategies. I copy below the abstract and a relevant quote (full paper
here: https://github.com/gr-g/ln-steady-state-model). In particular you
can find a link between the idea of "drain" you defined and the concept
of "demand imbalance" in the paper.
Abstract:
/In this paper, we consider an idealized scenario in which the Lightning //
//Network (or any similar payment network) has scaled to the size and //
//volume of a self-sustained economy, meaning that the number of on-chain //
//transactions - including channel opening and closing - has become //
//negligible when compared to the number of off-chain transactions, and //
//payments continuously flow across a network with relatively stable //
//topology. We take this scenario to the extreme and model a network
where //
//the channels are fixed, so that payments form a completely closed //
//system, and where nodes have (on a long enough timescale) stable and //
//perfectly balanced incoming and outgoing payments (i.e. they spend //
//exactly what they earn). We call this scenario the "steady-state //
//economy" of the payment network.//
//
//We argue that in such scenario, in a network of n connected nodes, //
//there is a tendency towards a state where exactly n-1 channels have //
//perfectly balanced flows in the two directions ("self-balancing" //
//channels), while all other channels are either unused, or have a //
//permanent tendency towards imbalance: the channel balance accumulates
at //
//one end and the channel is only intermittently available in one //
//direction ("stuttering" channels). We note that the "self-balancing" //
//channels form a spanning tree of the network graph, which we call the //
//"core spanning tree" of the payment network.//
//
//We also try to derive some practical lessons from this idealized //
//scenario, hopefully providing some useful insight to node operators of //
//the current (embryonic) Lightning Network.//
//
//At the end of the paper, we provide some remarks on the more general //
//case in which nodes do not balance their income and expenses.//
/
From section 4:
/There is general consensus on the //fact that having a large fraction
of channels
not usable or barely usable in one////direction is not a healthy
predicament for the
network, and that some form of////channel management will need to be
practiced
by node operators involving a mix////of rebalancing and fee fine-tuning.
However,
one of the main takeaways of the////analysis of the steady-state model
is that the
network might have a tendency////to push////most////of the channels
(when not unused)
towards being chronically////unbalanced.//
//We wonder if these two tools (rebalancing and fee management) are really//
//enough to contrast the tendency toward imbalance. If not, it would be
appro-//
//priate to consider also other strategies to “work with the imbalances”
instead//
//of fighting them.////We refer, for example, to efficient low-latency
mechanisms//
//to signal when a channel becomes unusable in one direction, in order
to limit//
//the failure rate, together with a general robustness of the network
against a//
//pervasive and high-volume flow of information about channels that
switch from//
//being available to not available and vice versa (or that switch
between low fees//
//and high fees)./
Kind regards,
Gregorio
_______________________________________________
Lightning-dev mailing list
Lightning-dev@lists.linuxfoundation.org
https://lists.linuxfoundation.org/mailman/listinfo/lightning-dev
