Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Benchmarking GNNs Using Lightning Network Data

Rainer Feichtinger, Florian Grötschla, Lioba Heimbach, Roger Wattenhofer

TL;DR

The paper addresses the scalability and routing efficiency of the Bitcoin Lightning Network by treating it as a graph and benchmarking Graph Neural Networks on diverse prediction tasks. It constructs a data-rich benchmark from gossip-based LN topology, on-chain capacity, and privacy-related aspects, and evaluates multiple GNN architectures, including GCN, GAT, GIN, GraphSAGE, and GraphGPS, across node- and edge-level tasks. Key findings show that GNNs leveraging topological and neighborhood information outperform baselines and that edge features confer benefits for several tasks, though not uniformly across all tasks like link prediction. The benchmark provides a practical framework for comparing GNNs on large-scale real-world payment networks and paves the way for improved routing strategies and network analytics.

Abstract

The Bitcoin Lightning Network is a layer 2 protocol designed to facilitate fast and inexpensive Bitcoin transactions. It operates by establishing channels between users, where Bitcoin is locked and transactions are conducted off-chain until the channels are closed, with only the initial and final transactions recorded on the blockchain. Routing transactions through intermediary nodes is crucial for users without direct channels, allowing these routing nodes to collect fees for their services. Nodes announce their channels to the network, forming a graph with channels as edges. In this paper, we analyze the graph structure of the Lightning Network and investigate the statistical relationships between node properties using machine learning, particularly Graph Neural Networks (GNNs). We formulate a series of tasks to explore these relationships and provide benchmarks for GNN architectures, demonstrating how topological and neighbor information enhances performance. Our evaluation of several models reveals the effectiveness of GNNs in these tasks and highlights the insights gained from their application.

Benchmarking GNNs Using Lightning Network Data

TL;DR

The paper addresses the scalability and routing efficiency of the Bitcoin Lightning Network by treating it as a graph and benchmarking Graph Neural Networks on diverse prediction tasks. It constructs a data-rich benchmark from gossip-based LN topology, on-chain capacity, and privacy-related aspects, and evaluates multiple GNN architectures, including GCN, GAT, GIN, GraphSAGE, and GraphGPS, across node- and edge-level tasks. Key findings show that GNNs leveraging topological and neighborhood information outperform baselines and that edge features confer benefits for several tasks, though not uniformly across all tasks like link prediction. The benchmark provides a practical framework for comparing GNNs on large-scale real-world payment networks and paves the way for improved routing strategies and network analytics.

Abstract

The Bitcoin Lightning Network is a layer 2 protocol designed to facilitate fast and inexpensive Bitcoin transactions. It operates by establishing channels between users, where Bitcoin is locked and transactions are conducted off-chain until the channels are closed, with only the initial and final transactions recorded on the blockchain. Routing transactions through intermediary nodes is crucial for users without direct channels, allowing these routing nodes to collect fees for their services. Nodes announce their channels to the network, forming a graph with channels as edges. In this paper, we analyze the graph structure of the Lightning Network and investigate the statistical relationships between node properties using machine learning, particularly Graph Neural Networks (GNNs). We formulate a series of tasks to explore these relationships and provide benchmarks for GNN architectures, demonstrating how topological and neighbor information enhances performance. Our evaluation of several models reveals the effectiveness of GNNs in these tasks and highlights the insights gained from their application.
Paper Structure (15 sections, 9 figures, 6 tables)

This paper contains 15 sections, 9 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Benchmark Introduction
  4. Gossip Messages
  5. Data Collection
  6. Dataset
  7. Benchmark Evaluation
  8. Tor Classification
  9. Capacity Regression
  10. Base Fee Regression
  11. Proportional Fee Regression
  12. Maximum HTLC Regression
  13. Link Prediction
  14. Conclusion
  15. Appendix

Figures (9)

  • Figure 1: Channel Updates
  • Figure 2: Node Address
  • Figure 3: Tor Classification
  • Figure 4: Capacity Distribution
  • Figure 5: Capacity Regression
  • ...and 4 more figures