基于队列的处理器亲和机制及在网络转发中的应用  

Queue-Based Processor Affinity and Its Network Forwarding

作　　者：姚崎[1] 韩臻[1] 高志明[1] 

机构地区：[1]北京交通大学计算机与信息技术学院,北京100044

出　　处：《北京交通大学学报》2010年第2期106-110,共5页JOURNAL OF BEIJING JIAOTONG UNIVERSITY

基　　金：国家"863"计划项目资助(2007AA01Z410);电子信息产业发展基金项目资助(2007-329)

摘　　要：提出一种新的基于多报文队列的处理器亲和机制,与传统基于网络接口的亲和机制相比,消除了多处理器对共享资源的竞争,优化了处理器的CACHE行为,降低了互斥锁的影响范围,能够有效提升多核处理器系统的网络转发处理能力.试验结果显示,在复杂网络环境下,基于队列的亲和机制报文转发性能较传统基于接口的亲和机制提高了1.8倍.A new affinity scheme based packets queues was proposed.Against the traditional affinity scheme based network interface,the new scheme eliminated the contention of shared resources,optimized the processor CACHE behavior,and reduced the impact of spin-lock.So the packet forwarding capability of multi-processors system was effectively raised.The experiment results shows that the performance of network packets forwarding with the new scheme can be improved 180 percent in the complex network environment.

关 键 词：网络转发 多核处理器 处理器亲和 CACHE行为 

分 类 号：TP393.03[自动化与计算机技术—计算机应用技术]