A Join Optimization Method for CPU/MIC Heterogeneous Systems

摘要

In recent years, heterogeneous systems consisting of general CPUs and many-core coprocessors have become the main trend in the high-performance computing area due to their powerful parallel computing capabilities and superior energy efficiencies. Join is one of the most important operations in database system. In order to effectively exploit each hardware's advantages in heterogeneous systems, in this paper we focus on how to optimize the join algorithm in hybrid CPU/MIC system. We design a join method with CPU and MIC working collaboratively when implementing the join operation. In order to fully utilize the MIC's parallel computing power, we also propose a Sort-Scatter-Join (SSJ) algorithm for MIC to generate the join index. Through turning the traditional process of comparison and matching into the process of computing and scattering, the SSJ gains more beneficial from thread-level parallelism and SIMD data parallelism. Experiment results show that, compared with the traditional parallel sort-merge join algorithm, the peak performance of the SSJ running on MIC is improved by around 26 %.