Performance Optimization of Top-k Queries on GPU

Abstract

With the development of web search engines, the concern on real-time performance of Top-k queries has attracted more and more attention. The author studies implement of classic algorithm No Random Access Algorithm in order to optimize performance of Top-k queries on GPU. We give a novel GPU algorithm by using the features of CUDA's programming model. Experiment results show that an implementation of the algorithm on one GPU runs more than 7000 times faster than a single core implementation on a latest CPU.