Performance enhancement in hybrid SDN using advanced deep learning with multi-objective optimization frameworks under heterogeneous environments

Abstract

The growth of software-defined networking (SDN) enhances network strength and provides flexible routing, especially in heterogeneous environments. Hence, an efficient framework is required for recent networks. Recently, hybrid SDN with the restricted deployment of SDN switches has been integrated with a conventional network that provides improved communication performance compared to traditional SDN systems. However, the recent hybrid SDNs lack effective link protection and optimal routing when used with complex topologies. Hence, this study presents a novel deep learning–based hybridized multi-stacked autoencoder with the duo-directed gated recurrent unit (MSAE-DDGRU) for automatic link failure prediction in hybrid SDN. Moreover, a multi-objective zebra optimizer (MO-ZeO) is introduced to perform optimal routing by solving multiple routing constraints. The developed study is processed with the Python platform, and publicly available GEANT topology is utilized for the whole experimental process. Various assessment measures like accuracy, precision, sensitivity, packet loss, cost, maximum link utilization (MLU), policy violation rates (PVRs), packet delivery ratio (PDR), and delay are analyzed and compared with existing studies. The developed technique achieved an accuracy of 96%, precision of 92%, sensitivity of 93%, PDR of 99.4%, PVR of 0.0005, and delay of 1.2 s are obtained.