An Openflow Based Anycast Scheme In Optical-Eelectric Heterogeneous Network

More and more large-scale data center cluster networks are challenging current network control and management progressively. Classical static two-layer VLAN management cannot satisfy the request of network supplier and users. The drawback of data center network management may impact the computing performance of the data center. Focusing on the scenario with data center cluster interconnection and large-scale data center intra networking, this thesis manages the data center networking via partitioning VLAN. We mainly investigate benefit the soft-definite network technique in data center network, especially heterogeneous data center networking. The main contribution lies in1) the unified control method of heterogeneous data center networking under soft-definite network;2) Virtualizing and resource separation and considering cross-layer path decision;3) employ OpenFlow protocol proposing a anycast scheme for heterogeneous network.The specific outcomes are expressed as follows:1. Traditional anycast service has many drawbacks, such as not allowing free access without restriction, hard to maintain communications and the weights of the paths are too unified, etc. To solve those problem, based on OpenFlow, we propose an anycast scheme achieving cross-layer control and verify its feasibility in heterogeneous network. With the unified control in soft-definite network, the proposed anycast scheme can choose the most proper interaction approach based on the network transportation, application layer requirement and data center resource. This scheme optimizes the network resource from the aspect of networking and cross-layer data transport path. This scheme realizes more dimension control in a circuit switching network and schedules the transport resource flexibly at the same time. In this way, the scheduling in heterogeneous network becomes more flexible, unified and effective. 2. The soft-definite network is short in heterogeneous network virtualizing and the joint scheduling of application resource and network resource. To overcome these, we propose a unified description method for lower layer resource in heterogeneous network. Besides, a network abstract algorithm and a dynamic feedback algorithm in virtual network are designed based on this unified description. The simulation results show that these two algorithms effectively match the service request and real-time network status. In addition, the unbalanced load of application resource has been improved dramatically. Hence, it is a feasible scheme for heterogeneous network virtualization.3. We improve the anycast scheme by integrating the soft-definite three-layer framework of network virtualization proxy. Based on classical anycast principles, the OpenFlow controller is extended. Besides, we design resource policy analyser, policy decision module, slice resource base and an abstract list of service. Furthermore, we improve the original FlowVisor, open the service connecter, design control policy module, path flash, service slice and path selection analyser, etc. Those changes enhances the feasibility of OpenFlow control framework in heterogeneous data center network. Based on those changes, we verify the small-scale flow and the feasibility of cross-layer control and anycast with OpenFlow.