Research On SDN Deployment And Their Applications In Data Centers

Since the emergence of Software Defined Networking (SDN), with the continuous development of science and technology and advance by academia and industry, the applications of Software Defined Networking are going on developing. SDN deployed in the data center is moving from theory to practice. This dissertation aims to study the Software Defined Networking deployment and applications in the data center.With the development of cloud computing, the data center has evolved into the distributed cloud data center. As we all know, nowadays data center is not only a center of many servers with high performance which could compute and store huge data, but also many geographically dispersed centers which across the wide area networks. The distributed cloud data center contains hundreds and thousands, even million of servers or PCs, which provide excellent support for computing. However, the development of distributed cloud data center also brings a series of new problems. Firstly, the large-scale and distributed servers and networks need centralized management, in order to improve the efficiency and availability of data center maintenance. Secondly, the distributed cloud data center carry the cloud computing business across the wide area networks, and its traffic engineering is also facing a severe challenge. The classification of elephant flow applications such as cloud computing, data backup directly affects the service quality of the data center. Again, with the development of the distributed characteristic, virtualization and other technologies, the security challenges of the distributed cloud data center network are facing the new situation.This dissertation analyzes and summarizes new characteristics of distributed cloud data center in deep, and focuses on three main issues:the controller placement, elephant flow detection, and DDoS detection issues of the distributed cloud data center in SDN. The main research contents of the dissertation are as follows:First of all, aiming at the lack of traditional data center networks, it focuses on the new features of distributed cloud data center. It analyzes three important issues that arise from the distributed topology, traffic engineering and network security. Aiming at the distributed topology and SDN controller placement issues of distributed cloud data center, we makes full consideration of the distributed characteristic, and present the new SDN domains partition based the spectral clustering and SDN controller placement algorithms, by which we can use the spectral clustering to partition a large network into several small SDN domains. In our algorithms, the matrix perturbation theory and eigengap are used to discover the stability of SDN domains and decide the optimal number of SDN domains automatically. To evaluate our algorithms, we develop a new experimental framework based on Cbench. The results show the effectiveness of our algorithm for the SDN domain partition and controller placement problems in distributed cloud data center, which provide important basis for the subsequent research of SDN application.Second, aiming at the traffic engineering issues in the distributed cloud data center and detecting the elephant flow, we propose an efficient elephant flow detection algorithm with Cost-Sensitive in SDN. To improve the real-time performance of the algorithm, we use two-stage flow detection strategy to classify traffic. A large number of mice flows are eliminated based on the head packet detection, which reduces the subsequent calculation, and then we apply C4.5 decision trees based on the flow characteristics to classify traffic. In order to further improve the algorithm's classification accuracy, we train the train data set with correlation analysis and introduce the cost-sensitive learning method to define a real-time elephant flow detection strategy, which effectively improve the classification accuracy of the algorithm and overcome the shortcomings of the existing algorithms. The experiments in SDN have been performed, showing that our strategy is good at detecting elephant flow.Finally, focusing on the problem of a new DDoS attack in the distributed cloud data center, namely link flooding attack, we propose an DDoS detection algorithm with Bloom Filter to deal with the link flooding attack. Our method is based on Bloom Filter and Software-Defined Networking, which can improve the real-time performance and efficiency of data storage. To save the time of extract packets and improve the real-time response, we scan the SDN flow table to get the statistics features of flows. In order to solve the problem of storage space and time efficiency, we use Bloom Filter to storage and detect the attacks. Compared with traditional detection methods, this method not only overcomes the problem of real-time data collection, but also solves the problem of massive data storage efficiency. Then we apply our method in SDN, extensive experiments show that our method is good at detecting the link flooding attack.