The SDN Research Based On Multi Controller

With the SDN network scale expansion, a single centralized control structure controller(such as NOX) processing capacity is limited, In the current expansion of the network can be a single centralized controller in two ways: one is through the controller to improve their ability to handle the way, another way is to use multiple controllers to improve the overall control of the processing power.Based on the above, we present the analysis of the two common architectural models, designed Zookeeper + RYU multi controller model, and the model to achieve a multi-cluster management controller, on the control plane to provide a more abstract user interface; a controller failure to cope with a single controller to avoid a single point of failure caused the problem, improve network reliability; control plane topology of the entire network to acquire, maintain consistency network status.According to the sender MAC address and destination port of the data stream as it matches the parameters, the initial allocation of the same data flow algorithm in the flow meter life cycle from the same MAC address by the same load balancing server processing, effectively reducing the packet loss rate. Based on MAC address instead of matching rules based on IP addresses, NAT is conducive to maintaining the state of the user during use, increasing the size identification algorithm to further improve the efficiency of the algorithm. On this basis, the proposed NAT address translation(Layer 3 forwarding) and MAC address translation(Frame Forwarding) combined traffic load balancing scheme, wherein the load balancing algorithm is statically allocated in part by a probability distribution principle to achieve traffic rights allocated by programs; dynamic adjustment algorithm for overloading the server load amortized through dynamic stretching full advantage of the range of the principle of each load balancing server resources, the current load progressively amortized and generate feedback chain, the load real equilibrium and continue to adjust, effectively load-balancing cluster of servers maintained in proper operating condition.