多Markov链预取模型在网络内存系统中的应用  

Applicedion of Multi-Markov Chains Prefetching Model in Network Memory System

作　　者：袁清波[1] 孙国忠[1] 陈明宇[1] 

机构地区：[1]中国科学院计算技术研究所计算机系统结构重点实验室,北京100080

出　　处：《计算机工程》2007年第23期105-107,共3页Computer Engineering

摘　　要：主机通过高速网络访问远程内存的性能已经达到或远高于访问本地磁盘的性能,通过各种优化手段,网络内存系统的性能能得到更好的提升。该文基于一个Linux网络内存系统(LNMS),在客户端一级提出了一种新的预取算法m-ppm,该算法发展了多Markov链预取模型,使之更适合LNMS。在LNMS上实现了另2种常用的预取算法以作比较,实验数据表明,m-ppm算法对多用户模式更有效。Good performance can be archived when fetching data from remote memory through high-speed network than that from local disk, and performance of network memory system can be improved by kinds of optimization. Based on a Linux network memory system（LNMS）, a new prefetching algorithm m-ppm is proposed to improve the performance in client part. The algorithm extends muhi-Markov chains prefetching model to make it more suitable for LNMS. Other two common prefetching algorithms are implemented in LNMS for comparison. Experimental results show that m-ppm method is more effective, especially in multi-user mode.

关 键 词：网络内存系统 多Markov链预取模型 性能优化 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]