Research On The Traffic Scheduling And Load Balancing Based On SDN In The Data Center Network

With the vigorous development of the Internet and cloud computing, the scale of the data center is growing, the business services in data center is complex and diverse, the network traffic rapid increase, Therefore, the underlying network management was put forward higher requirements. SDN as a new network paradigm, with programmable strong and has a global view of the network, more and more be deployed in the data center.This paper focuses on the data center load balancing and server cluster traffic scheduling, to take advantage of SDN centralized control, design the controller application, and improve the performance of the system.(1) By analyzing the relevant background and principle of traffic scheduling in data center found that, in the data center topology multipath widespread and a key limitation of traditional ECMP-based traffic scheduling scheme is that two or more large, long-lived flows can collide on their hash and end up on the same output port, creating an avoidable bottleneck.A SDN-based dynamic traffic scheduling for data center scheme is proposed to improve the utilization of the data center network. By combining Open Flow protocol and s Flow protocol to collect the network status, get link utilization and the bandwidth in real-time of the long-lived flows. Not only using the equal-cost path to avoid the congestion, also proposed a non-shortest path routing to make full use of link diversity. When the link utilization is too high on the shortest path, choose the non-shortest path to avoid the bottleneck link, in order to reduce the flow completion time. Finally, based on multi-commodity flow problem, the path selection of the elephant flow is modeled, and the global optimal solution is obtained by using the particle swarm optimization algorithm. Developed an application on Floodlight Controller to achieve a prototype system, the effectiveness of the system is verified by experiment.(2) This paper further studied the data center server cluster load balancing correlation method and proposed using Open Flow switches as a load balancer, through centralized scheduling of SDN controller, it can achieve the global optimum distribution request. The system combines Open Flow and s Flow protocol to collect network status and server load, in order to provide a centralized view to the controller for making a decision. The controller installs wildcard rules in the switches proactively to direct requests of large groups of clients without involving the controller which effectively saves the flow table space and reduces the delay of the network. Finally, write the controller applications implementation, carried out experiments in Mininet simulation environment. The preliminary evaluation results demonstrate that our dynamic load balancing scheme is superior to not only the random load balancing algorithm but also the round robin load balancing algorithm.