多核处理器环境下内存数据库索引性能分析  

The Performance Analysis of Main Memory Database Indices on Multi-Core Processors

作　　者：郭超[1] 李坤[1] 王永炎[1] 刘胜航[1] 王宏安[1] 

机构地区：[1]中国科学院软件研究所,北京100190

出　　处：《计算机学报》2010年第8期1512-1522,共11页Chinese Journal of Computers

基　　金：国家"八六三"高技术研究发展计划项目基金(2008AA04A105)资助~~

摘　　要：从20世纪80年代内存数据库出现时的T树到21世纪初出现的缓存感知的CSS、CSB+树等,都适应了当时的硬件发展趋势,具有一定的性能优势.随着计算机硬件技术的进一步发展,尤其是多核技术的应用与推广,新的多核处理器在提高索引性能的同时又给内存索引结构提出了新的挑战.文中对B+树、T树、CSS树、CSB+树等几个经典的内存索引结构在多核处理器环境下的性能进行了全面的实验测试,对其在多核处理器环境下不同数据输入、不同节点大小等多种情况下的性能构成与差异进行了比较和分析,总结了在多核处理器中影响索引性能的关键因素,为内存索引结构的进一步改进奠定了坚实的基础.There are more and more advanced technologies used to improve the performance of processors,e.g.,SMT and CMP.In one hand,these technologies improve the main memory indices＇ performance,and in the other hand,they create new challenges for these indices.To design a high-performance index,we should carefully evaluate the behavior of the multi-core processor while searching and updating.We have already known how traditional database indices perform on single-core CPU.This paper chooses some well-known indices including B＋tree,T tree,CSS tree,CSB＋ tree and provides a thorough experimental study to show how these indices perform on multi-core processors in different conditions,such as different node size,different data input,and so on.From the results,some useful advices about index improvement can be got.

关 键 词：内存索引结构 多核处理器 缓存感知 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]