A GPU Accelerated Temporal Window-Based Random Walk Sampler
This work addresses the scalability challenge of generating causality-preserving random walks from high-volume streaming temporal graphs, which is critical for real-time analysis in domains like microservices and finance.
Tempest is a GPU-accelerated engine for streaming temporal random walks that processes billion-edge streams under sliding windows in real time, outperforming prior systems in both ingestion and walk generation throughput while preserving causal correctness.
Temporal random walks, which sample causality-preserving paths, are widely used to analyze time-stamped interactions in domains such as microservices, finance, and online platforms. Generating such walks at scale is challenging because real-world graphs evolve as high-volume streams, making continuous ingestion, efficient memory usage, and strict temporal ordering essential for practical deployment. We present Tempest (TEMPoral nEtwork Streaming Traversals), a GPU-accelerated engine for streaming temporal random walks. Tempest combines a GPU-native dual-index organization over a shared edge store with a hierarchical cooperative scheduler that dispatches walks at thread, warp, or block granularity based on per-step node convergence, enabling efficient start-edge selection, hop-by-hop causality enforcement, and window-based eviction without synchronization. It further provides closed-form constant-time samplers for common temporal bias functions. Our evaluation demonstrates sustained real-time processing of billion-edge streams under sliding windows, outperforming prior systems in ingestion and walk generation throughput while preserving causal correctness.