高性能L5-Dispatcher的性能评测  

Evaluation on High Performance L5-Dispatcher

作　　者：雷迎春[1] 周幼英[2] 李军[3] 

机构地区：[1]中国科学院计算技术研究所国家智能计算机研究开发中心,北京100080 [2]浙江大学数学系,杭州310028 [3]华北水利水电学院网络中心,郑州450045

出　　处：《计算机研究与发展》2003年第3期475-482,共8页Journal of Computer Research and Development

基　　金：国家"八六三"高技术研究发展计划软件专项基金 ( 2 0 0 2AA1Z2 10 2 )

摘　　要：为支持基于请求内容调度的策略 ,提出并实现了一个集中调度、分布路由的可扩展路由机制 与通常的单前端机不同 ,提出的可扩展路由结构由多个前端机组成 一个前端机执行集中调度任务 (即调度器 ) ,其余的前端机执行分布路由任务 (即转发器 ) 测试数据表明 ,一台运行调度器程序的PⅢ 933主机的吞吐量为 4 5 0 0 0 0conns/s,可满足 80台转发器同时请求调度服务 换言之 ,有 80台转发器可同时转发请求数据或响应数据 ,即分布式前端总共可提供Presented and implemented in this paper is a scalable request routing mechanism for supporting scheduling policy based on request content, which is central scheduling and distributed routing Unlike common single front end, the scalable routing architecture is composed of several front ends One front end performs the central scheduling task, i e the dispatcher, while the others carry out distributed routing tasks, i e distributors The testing results show that the throughput of a PⅢ 933 host on which the scheduler runs is 450000 conns/s, capable of satisfying 80 distributors requesting scheduling service concurrently In other words, 80 distributors can distribute request or response at the same time, i e distributed front ends can provide throughput up to 16 Gbps

关 键 词：分发器 HTTP SPLICING 调度器 转发器 

分 类 号：TP393.05[自动化与计算机技术—计算机应用技术]