LBOSELM: SDN control plane load balancing using online sequential extreme learning machine

To manage software-defined networks (SDNs) efficiently, a centralized controller is essential for making network-wide decisions. In a distributed SDN, the task of distributing the load among controllers is challenging. However, the existing literature does not focus on real-time continuous traffic, which is important in today's internet world. There are situations in SDNs where new data arrives continuously, so frequent model updating with less computational cost is required. To alleviate this and to work on real-time data, we propose a load balancing using online sequential extreme learning machine, which uses an online sequential extreme learning machine (OSELM) for load balancing in the control plane. OSELM is an online learning technique for predicting the load of controllers in advance. It is an outstanding algorithm for modeling and predicting time series data, which is suitable for our controller load prediction. If any of the controllers is predicted as about to overload, the subsequent step is to balance the load of the control plane using the ameliorated technique, which uses a heuristic approach, before it gets overloaded. With this ameliorated approach, the migrated switch is selected, which has more impact on the overloaded controller and also the target controller. First, we implemented the OSELM algorithm on a dataset and got train and test scores as 74.5% and 73.1%, respectively. Next, we implemented the proposed technique in a mininet emulator to predict the load in advance and take appropriate migration to balance the load in the control plane using Ryu controllers. With our approach, we decreased the response time of the control plane in distributed SDNs compared to the existing literature to increase the performance efficiency of an SDN network.