The Optimization Of Multi-Join Query Based On GPU

With the development of the information society, data processing become more harder because of the more complex and huge amount of data. The serial processing of databases can not meet the need, parallel processing is the effective way to solve this problem. The development of the graphics processor with its powerful computing power and memory bandwidth, has become a powerful tool for parallel computing, and provide the condition for processing the data more quickly.Multi-join query is the most common and time-consuming operation in data processing, and the efficiency of multi-join query is an important factor of database performance. Therefore, this paper uses the GPU to optimize the multi-join operation. The optimization is divided into two stages, First, build a cost estimation of each join, and use the heuristic algorithm to build the minimum cost tree for the multi-join queries; The second stage is optimize the execution of the tree. The optimization is divided into intra and inter join. First, optimize the intra join, we analysis the serial implementations of the sort-merge join and the hash join, then we implement them in a parallel way. Second, we discuss the optimization of the inter join, which is the sequential, the layered parallel and the right deep tree execution policy, analysis their advantages and disadvantages. Finally, we test the sort-merge and hash join on GPU and multicore CPU platform. The result indicates that the performance on GPU is better than the performance on the multi-core CPU. The performance improvement can reach 7.25 X and 5.21 X. At the same time, we test the several different parallel execution policy on the GPU platform against the parallel algorithm based on the multi-core CPU. The result indicates that the multi-join algorithms based on GPU has a better performance, compared to their optimized CPU-based algorithms.This paper use the GPU to improve the performance of multi-join and get a certain effect, which can provide effective protection for improving the efficiency of database management.