Research On Optimizations Of TCP/IP Processing For Massive Network Traffic Application

In the rencent five years, the rapid development of internet makes Ethernet bandwidth growfrom 100Mbps to 1000Mbps, and then to 10Gbps. But processors are not keeping pace with itsperformance pace, particularly within the memory bandwi dth and access latency greatly lagsbehind the pace of technological advances in the development of network bandwidth. Thisgives real-time processing the massive network traffic serious challenges in many orientednetwork security system, which is used to access and monitor the information of internet. Inthis paper, we analyse the problems of Real-time Processing System of Massive NetworkTraffic, and then design TCP Acceleration Architecture (TAA), which is a new architecture ofprocessing TCP/IP protocol based on Intelligent Network Interface Card (INIC). In the TAA,we make INIC as the platform for capturing network packets, and implement partial protocoloffload on it in order to reduce the utilization of host system resource and enhance thethroughput of Real-time Processing System of Massive Network Traffic. The paper lucubrateson the network driver program of PPC440 which is an INIC, and does some experiments onimproved system. The experimental results show that the receiving throughput of PPC440 canreach line speed (1488Kpps).At the same time, the processing pattern of the packets on the INIC is studied to deal withpackets as much as possible without affecting the ability of capturing the packets, thus, toimprove the ability of Real-time Processing System of Massive Network Traffic. This paperdesigns and realizes two kinds of packets processing structure and makes in-depth analysis andcomparison of them. It is proved processing packets in the interruption is a better way. Whendoing limited processing of the packets, the capture performance of INIC could achieve linespeed (1488Kpps). This paper also proves the feasibility to implement partial protocol offloadon PPC440.