基于共享cache多核处理器的数据库内存排序优化

针对目前主流的多核处理器,提出了共享cache敏感的数据库排序多线程执行框架(sharedcache sensitive multithreaded sorting framework,SCS-MSF).首先分析了多线程QuickSort排序在共享cache多核处理器中执行时面临的性能瓶颈,在此基础上针对SCS-MSF每个处理阶段的数据访问特点,提出了各自的多线程并行执行模式,并通过各种优化策略改善线程执行时的cache性能,特别是减少多线程访问共享cache时的访问冲突问题,以提高线程的cache性能.在实验中,基于内存数据库EaseDB实现了SCS-MSF.实验结果表明SCS-MSF具有良好cache访问性能,从而提高了多线程执行的效率,而且性能稳定,数据库排序性能得到了较大提高.

Database Internal Sorting Optimization Based on Shared Cache Chip Multi-Processor

Based on chip multi-processor(CMP),a shared-cache sensitive muhithreaded sorting framework(SCS-MSF)is presented.SCS-MSF has three phases.Firstly,the performance bottlenecks are analyzed through an instance of multithreaded QuickSort execution in shared CMP.Then based on the analysis,aiming at the characters of data access in each phase,the execution patterns of multithread and optimization strategies are presented to improve their performance of cache access,especially to reduce the shared cache confliction.In the experiments,SCS-MSF is realized in the main-memory database EaseDB.The experimental results show that SCS-MSF has good cache behaviors and the efficiency of thread execution is improved.Futhermore,the performance is stable,which leads to a biggish performance improving of database internal sorting.