How Fast Can Graph Computations Go on Fine-grained Parallel Architectures

2025-07-01

AI Summary

Graph computation suffers from performance bottlenecks on conventional parallel architectures due to the irregular memory access patterns and severe load imbalance inherent in real-world graphs. To address these challenges, this paper proposes UpDown, a fine-grained programmable architecture co-designed across hardware and software to optimize graph traversal and iterative computation. UpDown supports efficient execution of multiple variants of key graph algorithms—including PageRank and BFS—while preserving high programmability. Evaluated on RMAT-generated graphs with 33 million processing elements, UpDown achieves 637K GTEPS for PageRank and 989K GTEPS for BFS, representing 5× and 100× speedups over prior state-of-the-art, respectively. By fundamentally rethinking architectural support for irregular workloads, UpDown establishes a new paradigm for scalable, high-performance graph processing.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Scalability, Parallel & Distributed Systems
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Economics, Online Markets and Human Computation: Incentives in network design for Web infrastructures and ecosystems

Abstract

Large-scale graph problems are of critical and growing importance and historically parallel architectures have provided little support. In the spirit of co-design, we explore the question, How fast can graph computing go on a fine-grained architecture? We explore the possibilities of an architecture optimized for fine-grained parallelism, natural programming, and the irregularity and skew found in real-world graphs. Using two graph benchmarks, PageRank (PR) and Breadth-First Search (BFS), we evaluate a Fine-Grained Graph architecture, UpDown, to explore what performance codesign can achieve. To demonstrate programmability, we wrote five variants of these algorithms. Simulations of up to 256 nodes (524,288 lanes) and projections to 16,384 nodes (33M lanes) show the UpDown system can achieve 637K GTEPS PR and 989K GTEPS BFS on RMAT, exceeding the best prior results by 5x and 100x respectively.

Problem

• Optimizing graph computations for fine-grained parallel architectures
• Enhancing performance of graph algorithms like PageRank and BFS
• Exploring scalability of UpDown architecture on large-scale graphs

Innovation

• Fine-grained parallel architecture for graph computing
• UpDown system optimized for irregular graph processing
• Achieves 5x-100x speedup on PageRank and BFS