The Technology Of Centrallized Management On Software Difined Network Cluster

The traditional network become increasingly complex and large. It can't meet the requirement of network resources flexible scheduling for business. Furthermore, due to the rapid development of business, network traffic surge, the demand for flexible control of traffic becomes more and more urgent. And also, such a huge network brings administrator tremendous pressure. In such a demand, SDN network architecture came into being. Its ideology of control plane separating from forwarding plane decouple easily. It not only solves the problem of resource scheduling flexibility, but also can program the network through the control plane. Administrator can also easily configure the network. But with the SDN network infrastructure applications, the control plane begin to become the bottleneck of network development. A single controller has been unable to meet demand, so the controller scale has become an important research project.Since the advent of the concept of SDN, the industry began to research the extension of controller and subsequently proposed several SDN cluster solutions. The solution of distributed database is more popular. While as a mainstream OpenDaylight controller also used this idea to realize its cluster. Since in the initial research stage of the cluster, its development is not mature and stable. In addition the function is not complete, especially for the lack of result about the principles and mechanisms of OpenDaylight cluster, however the demand for the cluster has been put on schedule.In this thesis, we conducted a comprehensive technical analysis of the OpenDaylight cluster, the cluster use AKKA technology, which provides two data channels for data and message synchronization between clusters. The data is stored using shard slice technology, while providing a file cache for data persistence. The cluster use the raft algorithm electoral mechanism for ownership in order to address master / slave election. In the southbound and northbound protocol plugin on the mechanism of cluster also line up analysis. OVSDB and OpenFlow protocol have been supported the clusters. The northbound RESTCONF is not particularly coping mechanisms, but without prejudice to the operation of the cluster. Although OpenDaylight has some support for cluster function, but for the southbound access and scalability of cluster has no further implementation mechanism.Therefore, this paper proposes a centralized control solutions based OpenDaylight cluster, the cluster aims to solve OpenDaylight southbound access and scalability issues. In order to enable flexible access to the cluster, we use OVSDB as device management protocol, and proposes a minimum connection load balancing algorithm to realize load balancing. It uses load factor to measure. We introduce the concept of wreck factor, in order to reduce the impact of the controller abnormality bringing to load balancing, which is used to measure the damage of the controller to load balancing. The wreck factor will continue to be adjusted while the device access the cluster; In addition, according to the adjustment of load and wreck factor to implement the dynamic expansion or volume reduction.