Study Of Key Technologies Of Network Processing Platform Based On Many-core

With the rapid development of the Internet,the network scale is growing more and more rapid,new network services are emerging day after day,and the structure of the new network system has been evolving.Huge network traffic,new network applications and network protocols have caused a new challenge to the packet processing performance and programming capability of network equipment.It requires that it not only has high speed packet processing capability,but also has the flexibility to adapt to the emergence of new network services and network protocols.The traditional ASIC based network equipment has been difficult to meet these two requirements,general CPU and NP based network devices are also existing defects in the performance and development cycle.This paper designs and implements the network processing platform based on the emerging technology – many-core,in order to achieve high-speed programmable network packet processing.The main work and innovation of this paper include:(1)NPPM(Network Processing Platform based on Many-core)is proposed.NPPM uses the clustering of the core organization structure,and through the load balanced packet scheduling technology and efficient co-processing technology to fully improve the efficiency of the cores,so as to improve the processing performance of the network processing platform.(2)The packet IO scheduling technology of NPPM is studied,and a packet IO scheduling technology based on the cyclic queue is proposed,which is called Cyc IO.CycIO balances load through circular queue,while reducing input packet delay;the packet scheduler support buffer management,to eliminate buffer management overhead with many-core,so as to enhance the many-core processing efficiency.(3)The co-processor technology of NPPM is studied,and a co-processor communication technology based on direct memory access is proposed,which is called DirectCop.DirectCop allows the co-processor to directly access the packet buffer of the core,and the co-processor can read and write the co-processing tasks by accessing the packet buffer of the processor core,effectively improve the communication efficiency between the core and the co-processor.(4)The prototype system of the 32-bit core network processing platform is implemented on Altera FPGA.The network processing platform contains 16 RISC processor cores,and each processor core runs on 100 MHz,which can achieve the performance of over 4Gbps line speed in programmable packet processing.Using the prototype system,the packet scheduling technology and co-processing technology are studied and analyzed.In summary,this paper makes a deep study on the network processing platform and its key technology,which has important reference value and practical value for the development and application of the network processing platform.