摘要
针对Lustre文件系统主要运用在高性能计算集群系统上的情况,分析Lustre文件系统的系统结构及其内部实现机制,测试系统在不同参数影响下的性能,认为Lustre可以运用在高性能集群系统上,也能运用在由普通计算机和TCP/IP网络构建的集群系统上。实验结果表明,在TCP/IP协议和千兆交换机下,Lustre的并发性很难发挥,但设置合适的客户端数和系统参数可以优化其性能,相对本地磁盘系统,Lustre系统仍有较大的性能提高和实用价值。
Lustre file system is mainly used in high‐performance computing cluster system .Its system structure and internal im‐plementation mechanism were analyzed ,and the system performance was tested through different parameters .Lustre can not on‐ly be used in high performance cluster system ,but in the cluster system built on the general use of computers and TCP/IP net‐work .Experimental results show that ,if Lustre is connected via TCP/IP protocols and gigabit switch ,its performance can be opti‐mized when the number of clients and some system parameters are properly set .Although its concurrency is depressed ,Lustre system still has large improvement on the performance comparing to local disk system .So it has practical value in this condition .
出处
《计算机工程与设计》
北大核心
2015年第6期1666-1670,共5页
Computer Engineering and Design
基金
国家自然科学基金项目(61370718)
教育部-中移动基金项目(MCM20130651)
广州市科技和信息化局基金项目(2014Y2-00006)
华南师范大学青年教师科研培育基金项目(2012KJ006)