Intelligent Routing in Software Defined Networking

Abstract

The explosion of the traffic and ascending in various network requirements newlinebecause of the tremendous increase in the networking and also multimedia data traffic newlinepossess numerous challenges in terms of routing complexity and scale of network newlinewhere traditional routing schemes are not appropriate for the Software-Defined newlineNetwork (SDN) because of their convergence limitations, adaptableness to changing newlinetopology of the network and lack of future vision based on the evolution of traffic in newlinethe network. Therefore, routing optimization and traffic engineering are pivotal to newlineadapting to these challenges, while maintaining the requirements of QoS. Accuracy in newlinethe estimation of traffic matrix is essential to solve various networking problems like newlinerouting, etc. SDN gives estimations of various flow types and hence opens up new newlineopportunities to tackle the problem of traffic matrix estimation. SDN provides the newlineglobal view through decoupling of data and control planes to provide efficiency in the newlinenetwork management and operation that are more beneficial to the traffic delivery based newlineon the policy, quicker routing response and helps in the reduction of investments and newlineexpenditures for operations of computing. Machine learning techniques have been used newlinewidely for solving the complex problems emerging in traffic engineering and routing newlineoptimization. At present days, the majority of the traffic predictions are based on ML newlinealgorithms, which offer better accuracy with respect to prediction. However, they newlinepointed out some shortcomings. In the proposed work, the DC-mLSTM-RNN based newlinetraffic prediction and traffic routing model for SDN. newlineIn the proposed DC-mLSTM-RNN based prediction model, the information newlineflows both forward and backward with recurrent cycles from the input nodes with the newlinenetwork and through the nodes of hidden passed into the nodes of output. Moreover, newlinethe model has different recurrent transition functions for each possible input, which newlinemakes it more expressive.