Design And Implementation Of SDN Switch Based On FPGA

In order to meet the increasingly complex network requirements,SDN(Software Defined Network)is gradually becoming a research hotspot.Its core technology,OpenFlow,separates the control plane of the network device from the data plane,which achieves flexible control of network traffic and provides a good platform for innovation of core networks and applications.At present,most SDN research is based on software simulation,which are hard to guarantee the performance for wire-speed processing.SDN research based on hardware platform is imperative.NetFPGA-1G is a research project at Stanford University,it is based on OpenFlow 1.0.The project in the article is based on the OpenFlow1.3 protocol,this project designs a switch hardware based on FPGA(Field-Programmable Gate Array).The switch is based on Xilinx Zynq-7000 SoC.The network flow processing acceleration is achieved on the PL(Program Logic)Part,and instructions forwarding is achieved on ARM processor.The hardware adopts 4 Gigabit ports.It realized 4Gbps hardware switching bandwidth and 4Gbps switching bandwidth with the control layer through DMA.Due to the limit logic resource,three-level flow table is set to build more traffic matching,and the parament of each flow table can be configured with registers.Through the serial port or configuring register,you can select different matching domains in each level flow table,such as source IP,target IP,and source port.At the same time,QoS system can optimize network performance.The switch designed in this paper basically realizes the content of OpenFlow1.3 protocol,and achieves 8Gbps network flow processing bandwidth.To verify the effectiveness of the hardware design,Iperf and real network traffic is used to test the functionality and performance of the switch.The test results show that the maximum rate of receiving and transmitting small-length packets is close to 800 Mbps,and the maximum rate of transmitting and receiving large-length packets is close to 900 Mbps.QoS system achieves 90% accuracy and is 20 times faster in data processing speed than packet processing with ARM Cortex-A9 dual core.