Density Estimations for Approximate Query Processing on SIMD Architectures

TL;DR

This paper explores using SIMD architectures like SSE and CUDA to significantly accelerate the computation of bandwidth in kernel density estimators for approximate query processing, enabling faster data exploration.

Contribution

It demonstrates the potential of SIMD architectures to drastically speed up bandwidth calculation in KDEs, improving AQP efficiency.

Findings

Orders of magnitude speedup over sequential implementations

Effective utilization of SSE CPU extensions and CUDA architectures

Potential for faster approximate query processing in large datasets

Abstract

Approximate query processing (AQP) is an interesting alternative for exact query processing. It is a tool for dealing with the huge data volumes where response time is more important than perfect accuracy (this is typically the case during initial phase of data exploration). There are many techniques for AQP, one of them is based on probability density functions (PDF). PDFs are typically calculated using nonparametric data-driven methods. One of the most popular nonparametric method is the kernel density estimator (KDE). However, a very serious drawback of using KDEs is the large number of calculations required to compute them. The shape of final density function is very sensitive to an entity called bandwidth or smoothing parameter. Calculating it's optimal value is not a trivial task and in general is very time consuming. In this paper we investigate the possibility of utilizing two SIMD architectures: SSE CPU extensions and NVIDIA's CUDA architecture to accelerate finding of the bandwidth. Our experiments show orders of magnitude improvements over a simple sequential implementation of classical algorithms used for that task.