Design And Implementation Of SDN Controller Cluster Based On Opendaylight

SDN (Software Defined Network) as a new type of Network architecture,including the advantages of separation of equipment with dataand Network traffic controlledlyflexibly, has been widely used in the data center and cluster service.However, with in-depth application of SDN network architecture,problems also gradually revealed. A single controller controlled all the resources in the network, once the controller failed,the service ability of network wouldreduce and even lost completely. In order to solve the above problem, this thesis presented a SDN controller cluster based on the opendaylight,the cluster was composed of many controllers which would avoid the problem of reliability and extensibility of the networkcaused by the failure of the single controller.Based on the home and abroad research about lateral extension of controller,combining the particularity of SDN network,this thesis summarized three main characteristics of the controller cluster which are the high availability, information consistency and access of cluster. In order to ensure the three characteristics,this paper implemented two aspects: the first aspect is the access of cluster, and this thesis mainly implemented the load balancing in the access of controller cluster. The second aspect is the notification of event. The realization of this function guaranteed information consistency between the upper business and controller cluster.(1) In this thesis, since the newest Openflow protocol supported one switch connected with many controllers,the master and slave controllers were determined by the Hash algorithm based on the MAC address of switch and election algorithm of master controller with the intention of balanceing and ensuring the control scope of each SDN controller.Moreover, once the controller was failed,the controller cluster could quickly recover from errors by the election algorithm of master controller designed by consideing about lots of factors.(2) In order to achieve the load balance function when the upper business accessing to the controller cluster,a HTTP load balance was deployed between the upper business and controller cluster. This thesis designed an adaptive Algorithm for Load Balancing by the time of controller processing the requirements on based of Ngnix polling policy, and the use of keepalived for the detection of failure ensuring the high availability of Ngnix server.(3) In order to ensuring the consistency of information between business module and controller cluster, this thesis achieved event notification of controller cluster based on the framework of AKKA cluster. The module captured the cluster events and notified the corresponding event based on the API provided by AKKA clustering and MD-SAL of Opendaylight. On the premise of the upper business module subscribed to the cluster events, it would accept the notification and then leaded to some business logic processing.Finally, all the functional modules were validated through constructing the testing environment. The results showed that the controller cluster designed in this thesis had higher availability, Information between the controller and the upper business kept consistent. The upper business and the underlying equipment accessing to cluster controller could achieve the goal of load balance.