用户空间协议栈的移植与性能优化

Transplantation and Performance Optimization of User-level Protocol Stack

文章将一个为嵌入式系统开发的轻量级用户空间协议栈LWIP移植到x86多核处理器上,并通过集成高速收发包引擎、优化关键数据结构及算法等措施提高其性能。论文介绍了与LWIP移植相关的系统模型设计、收发包引擎集成和操作系统抽象层实现,以及与LWIP优化相关的PCB管理、定时器管理与资源互斥锁实现。经实验,基于用户空间协议栈实现的web服务器,其性能比基于内核协议栈的web服务器提高了一倍。

In this paper, we transport a lightweight user-level protocol stack LWIP developed for embedded system to x86 multi-core processor, and improved its performance by integrating high-speed packet I/O engine and optimizing the key data structures and algorithms. This paper introduces the design of system model, the integration with high-speed packet I/O engine and the implementation of operating system abstraction layer related to the transplantation of LWIP, and the PCB management, the timer management and the implementation of resource mutex related to the optimization of LWIP. The final experiment shows that the performance of web server based on user-level protocol stack is two times of that of the web server based on kernel-level protocol stack.