Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bingo: Radix-based Bias Factorization for Random Walk on Dynamic Graphs

Pinhuan Wang, Chengying Huan, Zhibin Wang, Chen Tian, Yuede Ji, Hang Liu

TL;DR

Bingo addresses the challenge of efficient random walks on dynamic graphs by introducing a radix-based bias factorization that enables constant-time sampling and fast updates. The system employs an adaptive grouping strategy to dramatically cut memory while maintaining sampling fidelity, and leverages GPU-based parallelism to support high-throughput batched graph updates. Key contributions include the bias decomposition algorithm, streaming and floating-point bias handling, adaptive group representations, and a two-phase parallel delete-and-swap for updates, all validated against multiple real-world datasets with large performance gains. The work advances dynamic-graph sampling, enabling real-time graph learning, recommendations, and RAG tasks with substantial practical impact.

Abstract

Random walks are a primary means for extracting information from large-scale graphs. While most real-world graphs are inherently dynamic, state-of-the-art random walk engines failed to efficiently support such a critical use case. This paper takes the initiative to build a general random walk engine for dynamically changing graphs with two key principles: (i) This system should support both low-latency streaming updates and high-throughput batched updates. (ii) This system should achieve fast sampling speed while maintaining acceptable space consumption to support dynamic graph updates. Upholding both standards, we introduce Bingo, a GPU-based random walk engine for dynamically changing graphs. First, we propose a novel radix-based bias factorization algorithm to support constant time sampling complexity while supporting fast streaming updates. Second, we present a group-adaption design to reduce space consumption dramatically. Third, we incorporate GPU-aware designs to support high-throughput batched graph updates on massively parallel platforms. Together, Bingo outperforms existing efforts across various applications, settings, and datasets, achieving up to a 271.11x speedup compared to the state-of-the-art efforts.

Bingo: Radix-based Bias Factorization for Random Walk on Dynamic Graphs

TL;DR

Bingo addresses the challenge of efficient random walks on dynamic graphs by introducing a radix-based bias factorization that enables constant-time sampling and fast updates. The system employs an adaptive grouping strategy to dramatically cut memory while maintaining sampling fidelity, and leverages GPU-based parallelism to support high-throughput batched graph updates. Key contributions include the bias decomposition algorithm, streaming and floating-point bias handling, adaptive group representations, and a two-phase parallel delete-and-swap for updates, all validated against multiple real-world datasets with large performance gains. The work advances dynamic-graph sampling, enabling real-time graph learning, recommendations, and RAG tasks with substantial practical impact.

Abstract

Random walks are a primary means for extracting information from large-scale graphs. While most real-world graphs are inherently dynamic, state-of-the-art random walk engines failed to efficiently support such a critical use case. This paper takes the initiative to build a general random walk engine for dynamically changing graphs with two key principles: (i) This system should support both low-latency streaming updates and high-throughput batched updates. (ii) This system should achieve fast sampling speed while maintaining acceptable space consumption to support dynamic graph updates. Upholding both standards, we introduce Bingo, a GPU-based random walk engine for dynamically changing graphs. First, we propose a novel radix-based bias factorization algorithm to support constant time sampling complexity while supporting fast streaming updates. Second, we present a group-adaption design to reduce space consumption dramatically. Third, we incorporate GPU-aware designs to support high-throughput batched graph updates on massively parallel platforms. Together, Bingo outperforms existing efforts across various applications, settings, and datasets, achieving up to a 271.11x speedup compared to the state-of-the-art efforts.

Paper Structure

This paper contains 27 sections, 1 theorem, 7 equations, 17 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Dynamic Graph
  4. Random Walk Applications
  5. Monte Carlo Sampling Algorithms
  6. Motivation and Bingo Overview
  7. Bingo: Radix-based Bias Decomposition for Random Walk
  8. Bingo: Bias Decomposition Algorithm
  9. Bingo for Streaming Updates
  10. Bingo for Floating-Point Biases
  11. Complexity Analysis
  12. Bingo System Implementation and Optimizations
  13. Adaptive Group Representation
  14. Bingo's Parallel Batched Graph Updates
  15. Evaluation
  16. ...and 12 more sections

Key Result

theorem 1

(Correctness). Bingo ensures that the probability of choosing each neighbor remains the same before and after radix-based bias factorization, i.e., Equation (transition probability) holds for Bingo's sampling.

Figures (17)

  • Figure 1: Running example. Event $1$ contains one edge insertion and event $2$ one edge deletion.
  • Figure 2: Three classical Monte Carlo sampling methods for sampling on vertex $2$.
  • Figure 3: Bingo workflow. The right top is sampling while the right bottom is updating the graph.
  • Figure 4: Bingo on the running example.
  • Figure 5: Bingo insertion operation.
  • ...and 12 more figures

Theorems & Definitions (2)

  • definition 1
  • theorem 1