Cognitive virtual ad hoc mobile cloud-based networking architecture

Abstract

This thesis proposed cognitive techniques and intelligent algorithms that offered adaptive and advanced facilities to cloud-based networking by using Virtual Ad Hoc Mobile Cloud Computing Networks architecture (VAMCCNs). This is presented as a working case to address their global network challenges and to add cognitive support to the network design and implementation for better meeting traffic management and application requirements in mission objectives. The thesis concentrates on three main contributions. Firstly, an adaptive model, namely: a Heterogeneous Mobile Cloud Computing Network (HMCCN), was proposed to integrate different cloud networks architectures into one workflow. The cognitive data offloading task and the routing decision methods were applied using two different approaches: Fuzzy Analytic Hierarchy system (FAH) as a first approach and cognitive Software Defined Network (SDN) model as a second centralised approach. Experimental results show improvement in network reliability and throughputs, minimised in both nodes’ energy consumption and network latency with efficient intelligent data load balance and network resources allocation with best cloud model selection. Secondly, based on a virtual Ad Hoc cloud network with a realistic Random Waypoint Motion (RWM) model, an innovative cognitive routing algorithm was presented to improve efficient and reliable route selection among multiple possible routes. Routing protocols based on conventional, Fuzzy logic used important parameters with two data collections and decisions techniques and a new adaptive Intelligent Hybrid Fuzzy-Neural routing protocol (IHFN) that included prior knowledge to the network of the underlying motion and energy parameters were all proposed and compared. Results with the new hybrid algorithm shown a significant improvement to solve the network end-to-end performance degradation problem. The new hybrid protocol improved network throughput with an average of 20% higher than traditional Ad Hoc On-Demand Distance Vector (AODV) Routing protocol, improved the usage of network resources and reduced the maintenance process in adynamic topologies network. Finally, based on datasets collected from a realistic motion RWM model in a virtual Ad Hoc cloud network, the performance behaviour of six selected deep learning algorithms to predict the next steps of positions, speed and residual battery energy values of these mobile nodes have been evaluated and compared. This work goes further by presenting two algorithm's training techniques to predict the next 300-time steps of position, speed, and energy. Results and dissuasion show the differences concerning prediction accuracy between using the single node dataset model or Multiple node's dataset model.