Banar A. Methods and technologies for using artificial intelligence algorithms in software-defined networks under resource constraints

The dissertation is dedicated to the development and experimental substantiation of methods and technologies for applying artificial intelligence algorithms in software-defined networks (SDN) to enhance adaptive access control to limited resources. The primary focus is placed on ensuring the equitable distribution of access to scarce resources and maintaining the stable operation of the network infrastructure under constrained telecommunication network resource conditions. A method for adaptive application-layer limited-resource access control in a software-defined network is proposed, which, unlike approaches that alter general network characteristics, localizes the control action directly at the point of the actual resource deficit. The mechanism of intelligent access control to limited resources has been improved by introducing a specialized AI agent that generates control decisions considering telemetric features, load history, temporal characteristics of requests, the client's share in the total load, prior blocking events, and the number of active clients. The method for selecting a machine learning model for intelligent management in resource-constrained SDNs has been further developed, incorporating a comprehensive evaluation of models based on metrics of accuracy, decision stability, decision-making time, overhead costs, and energy consumption. The architecture of an intelligent SDN for operation under limited-resource conditions has been improved by integrating network, application, and IoT components into a single control loop and embedding the AI subsystem directly into the decision-making cycle. A Cloud-AI-SDN framework and an application web-service for management and monitoring have been developed and implemented, followed by hardware validation, which ensured the reproducibility of scenarios and the practical verification of the proposed solutions. The obtained results confirmed the efficiency of the proposed methods in enhancing adaptive access control to limited resources, ensuring an equitable distribution of access among clients, and maintaining the stable operation of the network infrastructure.