Options
REBAL: Channel Balancing for Payment Channel Networks
Journal
Proceedings - IEEE Computer Society's Annual International Symposium on Modeling, Analysis, and Simulation of Computer and Telecommunications Systems, MASCOTS
ISSN
15267539
Date Issued
2021-01-01
Author(s)
Awathare, Nitin
Suraj,
Akash,
Ribeiro, Vinay Joseph
Bellur, Umesh
Abstract
Cryptocurrency networks are a promising infrastructure for pseudonymous online payments. However, low throughput has prevented their widespread acceptance. A promising solution to scale throughput is the Payment channel network (PCN), exemplified by the Lightning Network (LN), that uses a network of off-chain bidirectional payment channels between parties that wish to transact often. Since payments use the shortest paths with sufficient funds over this network, channel balances get exhausted in the direction transactions flow and eventually become unidirectional. This results in transactions failing and consequently a lower transaction success ratio. Our observations on the production LN show that over 63% of the channels lose over 80% of the channel balance in one direction over time, which makes the success ratio of a real-world workload drop from 71% to 29%. A unidirectional channel along a path results in a failure message back to the source that recomputes the path, excluding the failed channel and reattempts the transaction, thus adding to the completion latency even for those transactions that do complete.We propose REBAL, a distributed re-balancing mechanism, and a new routing scheme to address the above issues. REBAL maximizes the extent to which channels can be re-balanced across the entire network. REBAL addresses the completion latency issue by re-routing transactions from intermediate nodes around a unidirectional channel rather than propagating the failure back to the source.Our comprehensive evaluation of REBAL shows that the success ratio improves from 30.18% to 79.54% and success volume from 3.98% to 29.99% for a real-world workload derived from the Ripple network, without adversely impacting the transaction latency. Even at very high transaction rates, REBAL outperforms Lightning Network Daemon (LND- a Golang implementation of LN) (12%) with a success ratio of 43.76%.