A GPU Accelerated Temporal Window-Based Random Walk Sampler

2026-05-15

AI Summary

This work addresses the challenge of efficiently generating causally consistent random walks over large-scale temporal graph streams under stringent throughput, memory efficiency, and temporal ordering constraints. The authors propose Tempest, a GPU-accelerated system that performs streaming temporal random walk sampling within a sliding time window. Tempest introduces three key innovations: a GPU-native dual-index structure over shared edge storage, a hierarchical cooperative scheduling mechanism that dynamically adjusts granularity based on node convergence degree, and a constant-time closed-form sampler supporting common temporal bias functions. The system enables efficient, synchronization-free execution of initial edge selection, hop-by-hop causally valid traversal, and window eviction. Experiments demonstrate that Tempest achieves sustained real-time processing on billion-edge streams, significantly outperforming existing systems in both data ingestion and walk generation throughput while strictly preserving causal correctness.

Abstract

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.

Problem

• temporal random walks
• streaming graphs
• causality preservation
• GPU acceleration
• sliding window

Innovation

• GPU acceleration
• temporal random walks
• streaming graph processing
• causality preservation
• sliding window