Relax and don't Stop: Graph-aware Asynchronous SSSP

Paper i proceeding, 2025

The Parallel Single-Source Shortest Path (SSSP) problem has been tackled through many implementations, yet no single approach consistently outperforms others across diverse graph structures. Moreover, most implementations require extensive parameter tuning to reach peak performance. In this paper, we introduce the AdaMW scheduler, which dynamically selects between the schedulers Wasp and Multi-Queue. To achieve this, we use graph sampling and heuristics to select and configure the scheduler. In contrast to common state-of-the-art bulk-synchronous implementations, AdaMW is fully asynchronous, thus not needing to stop for global barriers. The resulting scheduler is highly competitive with the best manually-tuned, state-of-the-art implementations.