Review Summary: A Cryptoeconomic Traffic Analysis of Bitcoin’s Lightning Network

Review Summary
‡ A Cryptoeconomic Traffic Analysis of Bitcoin’s Lightning Network. [PDF]
Ferenc Béres (Hungarian Institute for Computer Science and Control), István András Seres (Eötvös Loránd University), András A Benczúr (Hungarian Institute for Computer Science and Control)
(‡ accepted for both conference and journal)

Paper summaries from the reviewers:
“The authors analyze the economic incentives in the Lightning Network by simulating the network performance. The simulation is based on publicly available snapshots of the current network. Some of the unknown network parameters such as traffic and paths chosen are estimated based on a relatively small number of observations from popular nodes in the network itself.”

“A look at Lightning Network data, including the graph, routing, economics of channels, and fees. One finding of interest is that though the network is growing and has traffic, the large nodes which route many payments aren't making significant profit. The paper is very thorough, well written, with understandable charts and figures.  The traffic simulator gives a good guess as to what payments are occurring on the network and an idea of the fees being generated.”

“The author has designed a LN traffic simulator which enables him to study the economy of transaction fees in the Lightning network. He then has run experiments based on real data taken last year between February and March. His simulations involve different parameters such as the ratio of merchants in a random transaction. His article is more or less a statistical paper analyzing the results of his simulator and the impact of different parameters. It turns out that transaction fees are essentially underpriced. The author also adresses privacy concerns. Due to LN's topology and its small-world nature, a large fraction of payment routes involves less than 2 intermediaries. In this case, payments can possibly be deanonymized. To remedy, the author has designed an algorithm to add more nodes to traditional routing paths. This algorithm leads to a small increase of transaction fees that remains bearable.”

Comments on the strength of the paper:
“This was a thorough, easy to read analysis of the current crypto-economics of the Lightning Network.  It brought up plenty of interesting questions about the future of the network and behaviour of users. It also shows one of the big questions of LN: will routing fees be significant, and establish a fee market similar to on the Bitcoin blockchain?  Or will altruistic nodes persist?”

“To my knowledge, the author has designed the only LN traffic simulator that permits to analyze  transaction fees. The author claims that payment routing is not economically viable. However, the cost for running a node is not analyzed. For the sake of this analysis, the author has considered that all payments have the same value which is a strong hypothesis.”

“The article is consistent with another public study done by Bitmex.”

“The paper attempts to address a number of timely issues, and provides a solid foundation for future research into how the network's performance could be improved. To the best of my knowledge it is the first reasonable extrapolation of global parameters, such as real traffic and routing behavior based on the few publicly available numbers. ”

“The findings on payment privacy confirm an issue that has long been speculated about, and has already been addressed both in the specification as well as some implementations for quite some time (at least as early as February 2018). Quantifying the importance of these defensive techniques and evaluation of their effectiveness is an important contribution.”

Critiques & author responses:
“The author does not consider the fact that a transaction can now be split in different parts using AMP with c-lightning 0.8.0.”

“Well done analysis of the Lightning Network.  It's interesting that of LN nodes, the "vast majority exhibit altruistic behavior" - but wonder about levels of confidence in your expectation "to see more maturity and a similar evolution in the LN transaction fee market in the future”. Of course, we don't know what's going to happen, but this remark in the beginning colored my reading of the rest of the analysis and I seemed to see altruism all over the place, and wonder if LN is inherently more altruistic, or at least one where costs get pushed out to externalities.”

Response: We added Figure 17. to depict the current immature state of the LN fee market by quantifying the fraction of nodes that apply default base fee and fee rate.

“It certainly has some deficiencies, for example if doesn't address the large number of nodes and channels that are non-public, however it is a first step towards a more quantifiable network, which could guide future improvements.”

“The paper does not mention the non-public part of the network, estimated by some to be more than 40% of the overall channels (and likely an even larger fraction of the node population). These nodes are mostly consumer nodes, and not merchants, so they likely would not skew the results too much, but a brief discussion would have been good.”

“The calculation of the annual RoI in Section 6 is a bit disingenuous, there is no relationship between routing nodes overprovisioning capacity and the actual use of the network. This is evidenced by the fact that a routing node can manipulate its RoI to be arbitrarily small by committing unreasonable amounts of funds to its channels, which is likely what LNBig is experiencing. A more demand-driven allocation of funds could result in saner fee to funds ratios. Further evidence of this is in the second part of Section 6 where the value and number of transfers is increased, without increasing capacities and the RoI improves.”

These reviewer insights were also met by explaining (in section 3.2) why we did not consider private payment channels in this work. A new experiment investigating the capacity overprovisioning of central router nodes was also implemented. The related results are shown in the new Figure 18-19 and Table 2.

“Beyond the pure fee-pressure to use short paths there is another factor that was not addressed by the authors: longer paths may considerably increase the probability of a payment failing, or getting stuck. Maybe that could be considered for future work?”

“The authors verified that the privacy issue exists only in a single implementation (lnd), whereas other implementations already have protections against these attacks in place. Quantifying their effectiveness would have added more color and could have provided better parameters for the mechanisms.”

“The paper provides descriptions of how the base data can be gathered, and provide the simulator code used to derive the measurements. The experiments are repeated a number of times, however confidence intervals and convergence behavior is not reported.”

“The paper could be strengthened by more active tests of payment routing and fees to give better ground truth for the simulator. The paper is analyzing a live network and while the authors aren't modifying the network or running well connected nodes themselves to generate the data, they do check their model with publicly posted information from large network nodes.”

“Without the same costs of entry (miners vs LN nodes) and network-wide limitations (limited block size vs, I guess, UTXO set size for LN which has no direct limit) I would actually be surprised if they follow the same path.”

“The economy of fees is highly dependent on the routing protocol. It would be interesting to see if the simulator can be adapted with alternative routing protocols. In particular, Equation (1) is no longer true for Ant routing. Another study with a non constant $\alpha$ would be interesting.”

Response: We addressed this comment by adding a new paragraph (in section 4) discussing possible extensions of the traffic simulator with regard to the simulated payment value alpha. Furthermore, references to Ant routing and atomic multipath payments (AMPs) were also added to the text.