摘要
在Linux系统中存在一个基本规则：接收网络数据的CPU负责处理网络数据，而一般来说每个CPU接收的数据量是不一样的，由此可能导致系统负载不均衡，甚至出现性能瓶颈，其根本原因在于Linux中的硬中断与软中断结合过于紧密，该文提出了一种软中断分发的理论，扩充了Linux对SMP体系的支持，有利于提高Linux对海量网络数据的处理性能。
In Linux, a basic rule is that CPU which receives packets from network should deal with those packets by itself. This may result in the imbalance of system ' s load. Even one specific CPU may become the bottom-neck of performance. The main reason is the close affinity of hardirq and softirq of one CPU.This paper introduces a new theory about softirq-dispense which extend Linux ' s support for SMP and is beneficial to enhance the performance of network process of Linux.
出处
《计算机工程》
CAS
CSCD
北大核心
2003年第22期126-127,180,共3页
Computer Engineering
基金
国家重点基础研究规划"973"项目(G1999035806)
中国科学院知识创新工程重大项目(KJCX1-09)
