Proactive look-ahead control of transaction flows for high-throughput payment channel network

Wuhui Chen, Xiaoyu Qiu, Zicong Hong, Zibin Zheng, Hong Ning Dai*, Jianting Zhang

*Corresponding author for this work

Research output: Chapter in book/report/conference proceedingConference proceedingpeer-review

Abstract

Blockchain technology has gained popularity owing to the success of cryptocurrencies such as Bitcoin and Ethereum. Nonetheless, the scalability challenge largely limits its applications in many real-world scenarios. Off-chain payment channel networks (PCNs) have recently emerged as a promising solution by conducting payments through off-chain channels. However, the throughput of current PCNs does not yet meet the growing demands of large-scale systems because: 1) most PCN systems only focus on maximizing the instantaneous throughput while failing to consider network dynamics in a long-term perspective; 2) transactions are re-actively routed in PCNs, in which intermediate nodes only passively forward every incoming transaction. These limitations of existing PCNs inevitably lead to channel imbalance and the failure of routing subsequent transactions. To address these challenges, we propose a novel proactive look-ahead algorithm (PLAC) that controls transaction flows from a long-term perspective and proactively prevents channel imbalance. In particular, we first conduct a measurement study on two real-world PCNs to explore their characteristics in terms of transaction distribution and topology. On that basis, we propose PLAC based on deep reinforcement learning (DRL), which directly learns the system dynamics from historical interactions of PCNs and aims at maximizing the long-term throughput. Furthermore, we develop a novel graph convolutional network-based model for PLAC, which extracts the inter-dependency between PCN nodes to consequently boost the performance. Extensive evaluations on real-world datasets show that PLAC improves state-of-the-art PCN routing schemes w.r.t the long-term throughput from 6.6% to 34.9%.

Original languageEnglish
Title of host publicationSoCC '22: Proceedings of the 13th Symposium on Cloud Computing
PublisherAssociation for Computing Machinery (ACM)
Pages429-444
Number of pages16
ISBN (Print)9781450394147
DOIs
Publication statusPublished - 7 Nov 2022
Event13th Annual ACM Symposium on Cloud Computing, SoCC 2022 - San Francisco, United States
Duration: 7 Nov 202211 Nov 2022
https://dl.acm.org/doi/proceedings/10.1145/3542929

Publication series

NameProceedings of the Symposium on Cloud Computing

Conference

Conference13th Annual ACM Symposium on Cloud Computing, SoCC 2022
Country/TerritoryUnited States
CitySan Francisco
Period7/11/2211/11/22
Internet address

Scopus Subject Areas

  • Artificial Intelligence
  • Information Systems
  • Software
  • Computational Theory and Mathematics
  • Computer Science Applications

User-Defined Keywords

  • blockchain
  • deep reinforcement learning
  • graph neural network
  • payment channel network
  • transaction flow scheduling

Fingerprint

Dive into the research topics of 'Proactive look-ahead control of transaction flows for high-throughput payment channel network'. Together they form a unique fingerprint.

Cite this