GPU Accelerated Similarity Self-Join for Multi-Dimensional Data

TL;DR

This paper presents a GPU-accelerated algorithm for performing similarity self-joins on high-dimensional data, utilizing a grid-based index and various optimizations to improve performance and scalability on modern heterogeneous systems.

Contribution

It introduces a novel GPU-optimized self-join algorithm with specific indexing and pruning techniques, and demonstrates scalable performance on multi-GPU and distributed systems.

Findings

Outperforms state-of-the-art GPU self-join methods on real-world datasets

Effective data reordering enhances filtering power and reduces computations

Achieves scalable performance on multi-GPU and distributed-memory architectures

Abstract

The self-join finds all objects in a dataset that are within a search distance, epsilon, of each other; therefore, the self-join is a building block of many algorithms. We advance a GPU-accelerated self-join algorithm targeted towards high dimensional data. The massive parallelism afforded by the GPU and high aggregate memory bandwidth makes the architecture well-suited for data-intensive workloads. We leverage a grid-based, GPU-tailored index to perform range queries. We propose the following optimizations: (i) a trade-off between candidate set filtering and index search overhead by exploiting properties of the index; (ii) reordering the data based on variance in each dimension to improve the filtering power of the index; and (iii) a pruning method for reducing the number of expensive distance calculations. Across most scenarios on real-world and synthetic datasets, our algorithm outperforms the parallel state-of-the-art approach. Exascale systems are converging on heterogeneous distributed-memory architectures. We show that an entity partitioning method can be utilized to achieve a balanced workload, and thus good scalability for multi-GPU or distributed-memory self-joins.