GTS: GPU-based Tree Index for Fast Similarity Search (2404.00966v1)

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.

• The paper introduces a GPU-based tree index that leverages parallel processing and pivot-based pruning to significantly speed up similarity search.
• The paper presents innovative dynamic update strategies to efficiently manage both streaming and batch data without degrading query performance.
• The experimental results show GTS achieving up to 20x speed improvements over existing GPU methods and two orders of magnitude faster than CPU-based baselines.

GPU-Based Tree Index for Fast Similarity Search: An Evaluation of GTS

In recent years, similarity search has emerged as a critical operation in various fields such as multimedia retrieval, decision-making, and data visualization. This paper introduces a novel approach known as GTS, a GPU-based tree index designed to enhance the efficiency of similarity search by leveraging the computational power of GPUs. GTS aims to address the limitations of traditional CPU-based methods, which often struggle with high computational costs and an inability to handle high-throughput data management efficiently.

Contributions

GTS adopts a structured approach by integrating a pivot-based tree structure that allows efficient object pruning. It also employs list tables positioned to optimize GPU computing. The research primarily proposes two innovative features:

1. Parallel GPU Execution: The GTS index enables parallel processing of similarity search tasks across general metric spaces by effectively utilizing GPU capabilities. This parallel nature allows it to significantly outperform traditional CPU-based methods and even some GPU-optimized approaches by reducing unnecessary distance computations.
2. Handling of Dynamic Updates: The paper presents efficient update strategies catering to both streaming data and batch updates, thereby supporting dynamic scenarios without impacting query performance adversely.

Theoretical Contributions

The cost model proposed to evaluate search performance stands as a notable addition. This model establishes a balance between concurrency and pruning capabilities, crucial for optimizing similarity search performance. The application of the triangle inequality principle ensures a reduction in the search space, enhancing the computational efficiency.

Experimental Validation

The experimental results presented are robust and comprehensive, involving five real-life datasets, and demonstrate that GTS achieves efficiency gains by two orders of magnitude over the best-performing CPU-based baselines. Furthermore, GTS shows up to 20 times improvement in efficiency over contemporary GPU-based methods. This experimental validation underscores GTS’s efficacy in not only reducing computation time but also in streamlining data management in dynamic datasets.

Implications and Future Outlook

The implications of this work are far-reaching, with potential impacts on how large-scale dynamic data sets are managed. GTS offers an innovative framework for scaling databases, particularly vector databases, and handling complex queries that are both diverse and computationally intensive. The research lays the groundwork for potential extensions into approximate similarity search, possibly integrating learned indexes within the GPU architecture to improve speed further.

In conclusion, this paper contributes significant advancements in the field of GPU-accelerated similarity search. By addressing the challenges of parallelism and dynamic data update management, GTS provides a state-of-the-art solution for achieving high-efficiency, high-throughput similarity search. This work offers a scalable approach to data management in metric spaces, with practical applications across a range of data-intensive domains. Future research could investigate the integration of learned indices, which might offer approximations of distant metrics, further improving the computational efficiency on GPUs.