Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FlexiWalker: Extensible GPU Framework for Efficient Dynamic Random Walks with Runtime Adaptation

Seongyeon Park, Jaeyong Song, Changmin Shin, Sukjin Kim, Junguk Hong, Jinho Lee

TL;DR

Dynamic random walks resist precomputation due to runtime-dependent transition probabilities, creating a need for workload-generic GPU frameworks. FlexiWalker delivers this through two optimized kernels (eRJS and eRVS), a lightweight per-node cost model for runtime kernel selection, and compile-time workload specialization via Flexi-Compiler. The framework demonstrates substantial performance gains over both CPU and GPU baselines across multiple dynamic workloads and graphs, while maintaining low overhead and multi-GPU scalability. The work is open-sourced to facilitate adoption and future extensions.

Abstract

Dynamic random walks are fundamental to various graph analysis applications, offering advantages by adapting to evolving graph properties. Their runtime-dependent transition probabilities break down the pre-computation strategy that underpins most existing CPU and GPU static random walk optimizations. This leaves practitioners suffering from suboptimal frameworks and having to write hand-tuned kernels that do not adapt to workload diversity. To handle this issue, we present FlexiWalker, the first GPU framework that delivers efficient, workload-generic support for dynamic random walks. Our design-space study shows that rejection sampling and reservoir sampling are more suitable than other sampling techniques under massive parallelism. Thus, we devise (i) new high-performance kernels for them that eliminate global reductions, redundant memory accesses, and random-number generation. Given the necessity of choosing the best-fitting sampling strategy at runtime, we adopt (ii) a lightweight first-order cost model that selects the faster kernel per node at runtime. To enhance usability, we introduce (iii) a compile-time component that automatically specializes user-supplied walk logic into optimized building blocks. On various dynamic random walk workloads with real-world graphs, FlexiWalker outperforms the best published CPU/GPU baselines by geometric means of 73.44x and 5.91x, respectively, while successfully executing workloads that prior systems cannot support. We open-source FlexiWalker in https://github.com/AIS-SNU/FlexiWalker.

FlexiWalker: Extensible GPU Framework for Efficient Dynamic Random Walks with Runtime Adaptation

TL;DR

Dynamic random walks resist precomputation due to runtime-dependent transition probabilities, creating a need for workload-generic GPU frameworks. FlexiWalker delivers this through two optimized kernels (eRJS and eRVS), a lightweight per-node cost model for runtime kernel selection, and compile-time workload specialization via Flexi-Compiler. The framework demonstrates substantial performance gains over both CPU and GPU baselines across multiple dynamic workloads and graphs, while maintaining low overhead and multi-GPU scalability. The work is open-sourced to facilitate adoption and future extensions.

Abstract

Dynamic random walks are fundamental to various graph analysis applications, offering advantages by adapting to evolving graph properties. Their runtime-dependent transition probabilities break down the pre-computation strategy that underpins most existing CPU and GPU static random walk optimizations. This leaves practitioners suffering from suboptimal frameworks and having to write hand-tuned kernels that do not adapt to workload diversity. To handle this issue, we present FlexiWalker, the first GPU framework that delivers efficient, workload-generic support for dynamic random walks. Our design-space study shows that rejection sampling and reservoir sampling are more suitable than other sampling techniques under massive parallelism. Thus, we devise (i) new high-performance kernels for them that eliminate global reductions, redundant memory accesses, and random-number generation. Given the necessity of choosing the best-fitting sampling strategy at runtime, we adopt (ii) a lightweight first-order cost model that selects the faster kernel per node at runtime. To enhance usability, we introduce (iii) a compile-time component that automatically specializes user-supplied walk logic into optimized building blocks. On various dynamic random walk workloads with real-world graphs, FlexiWalker outperforms the best published CPU/GPU baselines by geometric means of 73.44x and 5.91x, respectively, while successfully executing workloads that prior systems cannot support. We open-source FlexiWalker in https://github.com/AIS-SNU/FlexiWalker.

Paper Structure

This paper contains 30 sections, 12 equations, 16 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Background and Analysis
  3. Dynamic Random Walks
  4. Analysis of Sampling Strategies
  5. Kernel Designs for Efficient Dynamic Random Walks
  6. Performance Comparison on Existing Sampling Methods
  7. Reservoir Sampling Optimization
  8. Rejection Sampling Optimization
  9. FlexiWalker Framework
  10. Flexi-Runtime
  11. Flexi-Compiler
  12. Implementation
  13. Flexi-Runtime Profiling Kernels.
  14. Concurrent Rejection and Reservoir Sampling.
  15. Dynamic Query Scheduling.
  16. ...and 15 more sections

Figures (16)

  • Figure 1: Single step of static and dynamic random walks.
  • Figure 2: An example of a random walk step and how different sampling methods select the next target node.
  • Figure 3: Performance comparison across various sampling methods. Execution time for (a) unweighted Node2Vec and (b) weighted Node2Vec is normalized to ITS (C-SAW).
  • Figure 4: Proposed optimizations of eRVS.
  • Figure 5: Optimizing rejection sampling with eRJS.
  • ...and 11 more figures