Research And Implementation Of High Speed Packet Processing Technology Based On X86 Architecture

With the Internet penetrated into all aspects of people's lives, many of its problems have become increasingly prominent. On the one hand,the network needs more speed, on the other hand massive data generated by the network needs deal with to solve the optimization problem. To the network information producing, forwarding, absorption and network monitoring, packet processing is very important. The software packet processing does not go well in performance, especially when encountering the combination of multi-core platform and Linux operating system. There have been more and more research results for this,including Linux's lastest versions that have improved Linux's packet processing performance. PF_RING, DPDK and NET-MAP are all high-speed frameworks for packet processing with better performace.Obviously, they have more or less similarities and differences in technology and characteristics. Users have some difficult in the selection and judgment on the performance and applicability of the frameworks for their actual development. Some articles have made performance comparison of forwarding ability among DPDK, PF RING ZC and NET-MAP in multiple dimensions with packets received and then transmitted immediately. Generally, most researches did not consider the scene of pure packet receiving. In addition, it needs more performance characteristics of frameworks' comparison on NUMA platform. What's more, it seems lack of research on their multi-core expansion problems.This thesis discussed and analyzed the existing Linux packet processing mechanism and key technology for optimization, focused on the analysis of the PF_RING and DPDK optimization technology and their similarities and differences. Then the comparative study of PF_RING and DPDK carried out. Firstly, it briefly explained the shortcomings of the existing research, and introduced Run-to-completion and Pipeline model (Style) classification. It proposed for that comparative study of different framework needs starting from two model and select several simplified model analysis.This thesis proposes an idealized model for packet processing performace analysis and research: an idealized theory based on traffic resistance. With traffic resistance model, the performance of PF RING and DPDK in the simplified Run-to-completion and Pipeline ways, with different load, were analysised. Followly, with similar research, they compared in scalability on multi-core platform. Then this thesis expanded similar researches to the NUMA platform. This research concluded performance characteristics of different frameworks on NUMA platform,especially for the scalability cross NUMA node and performance mapped by threads distribution styles among NUMA nodes. And then, based on the existing optimization technology of packet processing, an offline playback system for recorded traffic designed, with the performance analysis and optimization on multiple parameters of the system.The article finally summarized and explained the shortcomings and future research prospects.