网络缓存的零拷贝优化  

Zero Copy Optimization for Web Cache

作　　者：周敬利[1] 王晓锋[1] 余胜生[1] 夏洪涛[1] 

机构地区：[1]华中科技大学计算机科学与技术学院,武汉430074

出　　处：《计算机科学》2005年第9期35-39,共5页Computer Science

基　　金：国家自然科学基金(60373088);国防研究基金(4131605)

摘　　要：数据的冗余拷贝是影响网络缓存软件性能的重要因素。本文详细分析了当前广泛使用的网络缓存程序Squid的数据拷贝流程,利用快速文件传输和数据流拼接两种策略在Linux平台上实现了Squid的零拷贝优化。快速文件传输是将文件数据从文件系统内核缓存直接发送到远程主机,数据流拼接则是在网络协议栈内核缓存之间直接交换数据,二者都能够减少数据在内核空间和用户空间流动带来的冗余拷贝。实验证明,经过优化的Squid在主机CPU占用率和请求响应时间等方面的性能均有显著提高。The poor performance of current Web cache attributes to redundant data copies significantly. Squid, which is well known as a widely used Web cache program, is analyzed in detail in this paper from the point of view of its data movement. Zero copy optimization to Squid is implemented based on two strategies： faster file transfer and data stream splicing. The former makes file data sent to remote host from kernel buffer of local file system directly, and the latter means that input and outgoing data can be exchanged between kernel buffers of protocol stacks smoothly. Both of them reduce data copies resulted from data movement between kernel space and user space. Experiments prove that the optimized Squid predominates over original version remarkably in both CPU utilities and mean request response time.

关 键 词：网络缓存 零拷贝优化 冗余拷贝 文件传输 数据流拼接 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]