LOADS: Load Optimization and Anomaly Detection Scheme for Software-Defined Networks

Abstract

The decoupling of control functionality from the forwarding devices to the control plane in Software-Defined Networks (SDN) provides a unique platform to design a programmable and reconfigurable network. A single controller due to its limited capacity and resources may not handle heavy load traffic generated from various smart devices. In order to handle this, multiple controllers need to be deployed at the control plane so as to ensure improved efficiency and scalability of the network. The data flow by the distributed controllers fluctuates frequently which results in an uneven load distribution amongst different controllers. So, to handle the aforementioned issues, in this paper, a Load Optimization and Anomaly Detection Scheme (LOADS) is proposed. Using LOADS , the probability of switch selection is determined using the following two factors (i) distance from the switch to the controller, and (ii) resource consumption ratio of the switch to its controller. Also, an IP flow-based network anomaly detection module has been designed to classify the traffic as malicious or normal. In order to address the network anomaly, the LOADS scheme uses Access Control Policies (ACPs) on the user's behavior in the network. The proposed scheme is evaluated on Mininet emulator using POX controller with datasets of Internet Topology Zoo from BTNorthAmerica zone. The performance analysis reveals that LOADS minimizes the average execution time by 6.74% and 20.64% as compared to the existing competitive schemes, Distributed Hopping Algorithm (DHA) and Elastic Distributed Controller (ElastiCon). Also, it helps in improving the overall migration cost and response time of each controller. The proposed LOADS scheme has the migration cost of 55.1 milliseconds as compared to the ElastiCon and DHA schemes alongwith the migration cost of 110 milliseconds and 140 milliseconds respectively. In addition to the migration cost, the response time of the proposed scheme is 32.8 milliseconds as compared to DHA and ElastiCon which takes almost 90 milliseconds and 78 milliseconds respectively.