摘要
针对天网Maze(一个P2P网络文件系统)[1]单点中央服务器负载瓶颈的问题,通过对网络和系统真实数据的量化分析,确立可优化的网络拓扑参数;采用日志统计分析的方法,归纳出用户的行为特征、节点特性,采用K-均值类聚方法建立了用户超级节点选举算法模型。综合考虑前面两个分析结论,提出了服务器超级节点配合用户超级节点的混合型超级节点架构设计。最后描述了架构实现中关键流程和模块,并采集新构架数据进行简单评测,达到了预期的改进效果。
In order to solve the problem that the single centralized server architecture of Tianwang Maze (a P2P network file system) becomes a system bottleneck, several improvable network topology parameters are determined through quantitative analysis based on real network and system data. By summarizing user behavior's characteristics and node properties through statistics analysis from log files, an algorithm based on K-mean clustering is proposed for user super-node selection. On top of the two prior conclusions, A design of hybrid super-node architecture with mixed server-side and user selected nodes is presented, The critical logic and key components are also introduced. After briefly evaluating the new system, the goal to improve the system load is achieved as expected.
出处
《计算机工程与设计》
CSCD
北大核心
2008年第14期3766-3770,3784,共6页
Computer Engineering and Design
基金
深圳市科技计划基金项目(QK200605)
深圳市南山区科技计划基金项目(K2006084)