The Research On Optimization Of Data Join Operation Based On GPU

With the booming development of information science technology and urgent marketrequirement, the performance of GPU is remarkably boosting. For the reason of differentcircuit design, the GPU has significant advantage over CPU both in computationalperformance and memory bandwidth. Having gone through the stages of fixed functionarchitecture, disjoint rendering architecture and unified rendering architecture, the capabilityof general parallel computation is gradually promoting. Thus we are able to accelerate SQLdatabase operations by making the most of it. For example, we can speed up data access withthe help of GPU parallel computing, and get faster data operation via graphic pipelines.This paper has studied several related SQL query algorithms, and mainly focused on theoptimization of join operations. And this paper present a new design of the three classicalrational join algorithms (nested-loop join, sort-merge join and hash join). We implementedthese algorithms both on multi-core CPU and on NVIDIA GPU.Fully integrated with the features of GPU, this paper has discussed the overall design ofthese three algorithms, and gave detailed analyses of their implementations.In chapter5, we tested these algorithms on multi-core CPU and GPU; the benchmarktests showed that the result sets running on GPU have an overall better performance. Fornested-loop, sort-merge and hash join, compared to their optimized CPU-based counterparts,the performance improvement can reach the peak value of9.0X,1.7X and2.3X.