基于遗传算法的网络处理器异构资源映射方法研究

Methodology of Mapping Network Processor Heterogeneous Resources Based on Genetic Algorithm

随着深亚微米工艺的迅速发展,现代网络处理器芯片广泛采用MPSoC(Multi-Processor System on Chip)体系结构实现,继而需要一种新的设计方法指导网络处理器体系结构设计.本文研究了网络处理器的设计方法,提出了一种基于遗传算法的网络应用到网络处理器异构硬件资源映射方法.该方法首先对网络处理器设计的问题空间进行分析,采用加权数据流进程网络描述网络应用,并参数化各种硬件资源,最后构建遗传算法来完成网络应用到异构硬件资源的映射,形成网络处理器体系结构设计方案.

As the fast development of deep sub-micron process, the implementation of modern network processors has been based on MPSoC （Multi-Processor System on Chip） architecture. New design methodology should be presented to guide the architectural design of network processors. In this paper, an efficient mapping method from network applications to network processor heterogeneous hardware resources is presented. This mapping method is based on genetic algorithms. Network applications are represented by Weighted Dataflow Process Network （WDPN）,heterogeneous network processor resources are described with system parameters ,and then the application model is mapped into the heterogeneous hardware resources to analyze the performance of network processors through the well-designed genetic algorithm ,and further to obtain the proper implementation scheme of NP architecture.