New algorithm solves global termination in distributed betweenness centrality

Computing betweenness centrality on large networks is computationally expensive, requiring aggregation of shortest-path dependencies across all vertex pairs. While scalable distributed algorithms exist, current implementations lack a lightweight system-wide termination detector. This oversight causes excessive messaging after local convergence or, if misconfigured, premature termination that corrupts results. The authors propose a global termination detection mechanism that enables each vertex to determine locally when the entire distributed computation has converged, eliminating the need for central coordination.

Using a custom Python simulator, the team compared their global approach against a local stopping strategy across synthetic graphs (Erdős–Rényi and geometric) and real-world networks (Email and Road networks). Results show that global termination detection lets vertices stop safely with zero final error across all evaluated topologies. Local stopping, meanwhile, leads to premature termination and measurable errors on heterogeneous networks. The findings underscore the importance of coordinated halting in distributed graph analytics, offering a practical solution for bandwidth-limited or very large-scale network environments.