Outperforming Dijkstra on Sparse Graphs: The Lightning Network Use Case

Abstract: Efficient routing is critical for payment channel networks (PCNs) such as the Lightning Network (LN), where most clients currently rely on Dijkstra-based algorithms for payment pathfinding. While Dijkstra's algorithm has long been regarded as optimal on sparse graphs, recent theoretical work challenges this view. The new Bounded Multi-Source Shortest Path (BMSSP) algorithm by Duan et al. theoretically achieves $O(m~log^{2/3}~n)$ runtime, which is asymptotically faster than Dijkstra's $O(m + n~log~n)$ on sparse directed graphs. In this paper, we implement BMSSP on Rust and compare its performance against Dijkstra's using real LN topology data. Our evaluation, based on multiple randomized trials and statistical tests, shows that current implementations of BMSSP do not significantly outperform Dijkstra's in practice, and speedups are smaller than what theory predicts, possibly due to implementation and constant factor overheads. These results provide the first empirical evidence of BMSSP's potential to accelerate LN routing and inform future optimizations of PCN pathfinding algorithms.