An empirical study of availability and reliability properties of the Bitcoin Lightning Network
The Bitcoin Lightning network is a mechanism to enable fast and inexpensive off-chain Bitcoin transactions using peer-to-peer (P2P) channels between nodes that can also be composed into a routing path. Although the resulting possible channel graphs are well-studied, there is no empirical data on the network's reliability in terms of being able to successfully route payments at a given moment in time. In this paper we address this gap and investigate two forms of availability that are a necessary ingredient to achieve such reliability. We first study the Lightning network's ability to route payments of various sizes to nearly every participating node, over most available channels. We establish an inverse relationship between payment volume and success rate and show that at best only about a third of destination nodes can be successfully reached. The routing is hampered by a number of possible errors, both transient and permanent. We then study the availability of nodes in the network longitudinally and determine how long-lived they are. Churn in the network is actually low, and a considerable number of nodes are hosted on cloud providers. By testing node liveness, we find that the propagated network information is relatively often stale, however, both for IP addresses and Tor onion addresses. We provide recommendations how the Lightning network can be improved, including considerations which trade-offs between privacy and decentralization on the one hand and reliability on the other hand should at least be reconsidered by the community developing the Lightning network.