SDN Based Traffic Management Study In Data Center

In recent years,with the development of cloud computing and big data,data draw great attention as an important resource.Data center,as a place storing and using this data,has been developed in scale and number.Different from the user-oriented service traditional data center provided,newly data center operating science computing service,and the traffic in data center is mostly east-west traffic between racks.Those traffic have high bandwidth,low latency and burst characteristic.How to manage those traffic to meet the development of new applications is a problem to be solved.Since Software Defined Network was introduced,SDN has drawn great attention for decoupling controlling plane and data plane,programmable and open interface.Integrating SDN into data center,make controlling plane get the global view of the network topology and make strategy flexibly.Data center independents on any device provider.OpenFlow as the first commercial SDN standard south interface,can apply to the real network.We propose ProActive Pod-Based scheduler based on SDN and apply it in the fat-tree topology.We analyze the fat-tree structure and extract a simplified network.We analyze and compare the fault of traditional algorithms and proposed a pod based load balancing algorithm,the key is to split traffic evenly to all the available paths and prevent consecutive packets forwarded to one node.We adopt proactive flow instantiation to reduce the communication overheads between controlling plane and forwarding plane,and pod-based flow entry design reduces the number of flow entries in OF switch.In this way,packet can be forwarded as a faster rate,decrease the data buffer in switch.Through this paper,we adopt the proactive pod-based flow instantiation in SDN and fat-tree based datacenter.We found that our scheduler can reduce the network congestion greatly and reduce the unbalancing rate of network thus get a high throughput and low latency.