电子学报 ›› 2009, Vol. 37 ›› Issue (2): 404-409.

• 论文 • 上一篇    下一篇

数据流窗口连接查询处理器研究

钱江波1, 王永利2, 陈 征1, 陈华辉1, 金 光1   

  1. 1. 宁波大学信息科学与工程学院,浙江宁波 315211;2. 南京理工大学计算机科学与技术学院,江苏南京 210094
  • 收稿日期:2007-08-01 修回日期:2008-10-15 出版日期:2009-02-25 发布日期:2009-02-25

Hardware Processor for Window Joins over Multiple Data Streams

QIAN Jiang-bo1, WANG Yong-li2, CHEN Zheng1, CHEN Hua-hui1, JIN Guang1   

  1. 1. School of Information Science and Engineering,Ningbo University,Ningbo,Zhejiang 315211,China;2. School of Computer Science and Technology,Nanjing University of Science and Technology,Nanjing,Jiangsu 210094,China
  • Received:2007-08-01 Revised:2008-10-15 Online:2009-02-25 Published:2009-02-25

摘要: 高速数据流处理是数据流管理系统的一个关键问题.目前已有系统一般采用查询优化、系统调度、降载等方法来提高速度,在高速数据流环境下存在明显的不足.为了最大程度地提高数据流连接操作的速度,提出专用硬件处理器WJSP及设计.针对千变万化连接条件所共同的基本操作,提出在WJSP可执行的指令系统WJSI,并提出多核处理器以提高WJSP的扩展性和并行性.实验结果显示WJSP原型比STREAM的速度提高10多倍,具有相当高的处理性能.WJSP可嵌入到路由器、交换机、传感器等设备中,提供高速数据流连接处理.

关键词: 连续查询, 窗口连接, 查询处理, 指令集设计

Abstract: Accelerating data processing is one of the key problems in DSMS(Data Streams Management System)research.Many researchers promote processing speed by query optimization.It potentially results in saturating the CPU.Load shedding is a candidate choice when a DSMS is executing aggregate operations and it becomes overload.However,load shedding can not be applied on join operation for it will potentially lose many results.In order to greatly accelerate join processing,we present a novel hardware processor implementation for window join evaluation over multiple data streams.To address variations in join conditions,we propose a set of instructions that can be executed in the processor.In addition,we introduce a multi-core processor to improve scalability and flexibility of the processor.Experimental results show that the processor outperforms software by more than ten times.

Key words: continuous query, window joins, query processing, instruction-set design

中图分类号: