Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GTS: GPU-based Tree Index for Fast Similarity Search

Yifan Zhu, Ruiyao Ma, Baihua Zheng, Xiangyu Ke, Lu Chen, Yunjun Gao

TL;DR

GTS presents a GPU-optimized, pivot-based tree index for fast, exact similarity search in general metric spaces without coordinate information. By storing indexing data in a table-backed layout and employing a two-stage, memory-aware batch search, it achieves high throughput while avoiding GPU memory deadlocks; it also supports streaming and batch updates with a reconstruction-based strategy. A cost model guides the trade-off between pruning and parallelism, and extensive experiments show up to two orders-of-magnitude gains over CPU baselines and up to 20x gains over prior GPU approaches across five real datasets. The work has practical impact for scalable similarity search in dynamic, large-scale domains, including diverse omics and multimedia datasets, and points toward future integration with learned or approximate GPU indices.

Abstract

Similarity search, the task of identifying objects most similar to a given query object under a specific metric, has gathered significant attention due to its practical applications. However, the absence of coordinate information to accelerate similarity search and the high computational cost of measuring object similarity hinder the efficiency of existing CPU-based methods. Additionally, these methods struggle to meet the demand for high throughput data management. To address these challenges, we propose GTS, a GPU-based tree index designed for the parallel processing of similarity search in general metric spaces, where only the distance metric for measuring object similarity is known. The GTS index utilizes a pivot-based tree structure to efficiently prune objects and employs list tables to facilitate GPU computing. To efficiently manage concurrent similarity queries with limited GPU memory, we have developed a two-stage search method that combines batch processing and sequential strategies to optimize memory usage. The paper also introduces an effective update strategy for the proposed GPU-based index, encompassing streaming data updates and batch data updates. Additionally, we present a cost model to evaluate search performance. Extensive experiments on five real-life datasets demonstrate that GTS achieves efficiency gains of up to two orders of magnitude over existing CPU baselines and up to 20x efficiency improvements compared to state-of-the-art GPU-based methods.

GTS: GPU-based Tree Index for Fast Similarity Search

TL;DR

GTS presents a GPU-optimized, pivot-based tree index for fast, exact similarity search in general metric spaces without coordinate information. By storing indexing data in a table-backed layout and employing a two-stage, memory-aware batch search, it achieves high throughput while avoiding GPU memory deadlocks; it also supports streaming and batch updates with a reconstruction-based strategy. A cost model guides the trade-off between pruning and parallelism, and extensive experiments show up to two orders-of-magnitude gains over CPU baselines and up to 20x gains over prior GPU approaches across five real datasets. The work has practical impact for scalable similarity search in dynamic, large-scale domains, including diverse omics and multimedia datasets, and points toward future integration with learned or approximate GPU indices.

Abstract

Similarity search, the task of identifying objects most similar to a given query object under a specific metric, has gathered significant attention due to its practical applications. However, the absence of coordinate information to accelerate similarity search and the high computational cost of measuring object similarity hinder the efficiency of existing CPU-based methods. Additionally, these methods struggle to meet the demand for high throughput data management. To address these challenges, we propose GTS, a GPU-based tree index designed for the parallel processing of similarity search in general metric spaces, where only the distance metric for measuring object similarity is known. The GTS index utilizes a pivot-based tree structure to efficiently prune objects and employs list tables to facilitate GPU computing. To efficiently manage concurrent similarity queries with limited GPU memory, we have developed a two-stage search method that combines batch processing and sequential strategies to optimize memory usage. The paper also introduces an effective update strategy for the proposed GPU-based index, encompassing streaming data updates and batch data updates. Additionally, we present a cost model to evaluate search performance. Extensive experiments on five real-life datasets demonstrate that GTS achieves efficiency gains of up to two orders of magnitude over existing CPU baselines and up to 20x efficiency improvements compared to state-of-the-art GPU-based methods.
Paper Structure (19 sections, 2 theorems, 3 equations, 11 figures, 5 tables, 5 algorithms)

This paper contains 19 sections, 2 theorems, 3 equations, 11 figures, 5 tables, 5 algorithms.

Table of Contents

  1. Introduction
  2. Related Work
  3. Problem Formulation
  4. Index
  5. Overview
  6. Index Structure
  7. Index Construction
  8. Index Updating
  9. Complexity Analyses
  10. Similarity Search
  11. Metric Range Query in Batch
  12. Metric kNN Query in Batch
  13. Cost Model
  14. Experiments
  15. Experimental Settings
  16. ...and 4 more sections

Key Result

lemma 1

Given a pivot $p$, a query object $q$, and a search radius $r$ in a metric space, an object $o$ can be pruned if $\vert d(o,p) - d(q,p)\vert > r$.

Figures (11)

  • Figure 1: Metric space and pivot mapping
  • Figure 2: The overall framework of our work GTS
  • Figure 3: The structure of our proposed GPU-based tree index, GTS
  • Figure 4: Illustration of the filtering process
  • Figure 5: Update cost
  • ...and 6 more figures

Theorems & Definitions (4)

  • definition 1
  • definition 2
  • lemma 1
  • lemma 2