Research And Implementation Of Software Load Balancing Method Based On Asymmetric RSS

The growing of network traffic makes the load balancing system become an essential component in the data center network.A flexible and high-performance load balancing system,as an entry point to distribute traffic,has major impact on the performance and availability of the network.When load balancing and address translation technologies work together,the asymmetric sessions bring challenges to the design of load balancing methods.From the perspective of dispatching data packets of asymmetric session and analyzing of the temporal and spatial features of packet sequences,a software load balancing method,which improves equipment resource utilization and network throughput,is proposed.Meanwhile,a software load balancing system is designed and implemented based on this method.This thesis consists of two parts:Firstly,a software load balancing method based on asymmetric RSS is proposed.This method dispatches the data packets to the expected CPU core,by analyzing the packet hashing mechanism of the network card.In order to offload the process of elephant flows,an elephant flow recognition model is constructed by mining the features of the sessions.Secondly,a software load balancing system based on asymmetric RSS is designed and implemented,including hardware abstraction layer,data forwarding layer and session control layer.This system can dispatch network packets to appropriate CPU cores,select service nodes according to load balancing strategies,create address mappings,identify network flow features,and deliver the information of elephant flows to the hardware flow table in the network card.The experimental results show that this system can dispatch data packets to the correct CPU core and can accurately identify and offload the elephant flow among sessions under different number of sessions and different ratios of elephant flow.On the same time,this system can make full use of the multi-core processor and the network card and effectively improves the packet processing throughput rate of load balancing system.