Doctor of Philosophy
Electrical and Computer Engineering
In this thesis, we address some of the challenges that the Intelligent Networking Automation (INA) paradigm poses. Our goal is to design schemes leveraging Machine Learning (ML) techniques to cope with situations that involve hard decision-making actions. The proposed solutions are data-driven and consist of an agent that operates at network elements such as routers, switches, or network servers. The data are gathered from realistic scenarios, either actual network deployments or emulated environments. To evaluate the enhancements that the designed schemes provide, we compare our solutions to non-intelligent ones. Additionally, we assess the trade-off between the obtained improvements and the computational costs of implementing the proposed mechanisms.
Accordingly, this thesis tackles the challenges that four specific research problems present. The first topic addresses the problem of balancing traffic in dense Internet of Things (IoT) network scenarios where the end devices and the Base Stations (BSs) form complex networks. By applying ML techniques to discover patterns in the association between the end devices and the BSs, the proposed scheme can balance the traffic load in a IoT network to increase the packet delivery ratio and reduce the energy cost of data delivery. The second research topic proposes an intelligent congestion control for internet connections at edge network elements. The design includes a congestion predictor based on an Artificial Neural Network (ANN) and an Active Queue Management (AQM) parameter tuner. Similarly, the third research topic includes an intelligent solution to the inter-domain congestion. Different from second topic, this problem considers the preservation of the private network data by means of Federated Learning (FL), since network elements of several organizations participate in the intelligent process. Finally, the fourth research topic refers to a framework to efficiently gathering network telemetry (NT) data. The proposed solution considers a traffic-aware approach so that the NT is intelligently collected and transmitted by the network elements.
All the proposed schemes are evaluated through use cases considering standardized networking mechanisms. Therefore, we envision that the solutions of these specific problems encompass a set of methods that can be utilized in real-world scenarios towards the realization of the INA paradigm.
Summary for Lay Audience
Imagine living in a huge city where the traffic of the vehicles is solely controlled by officers: no traffic lights, no barricades, no separators, no signage, just traffic officers. What if a massive event is taking place near your home and you did not know? What if an accident occurs on a road you just merged onto? It is hard to visualize the flows of the vehicles going smoothly. Although the city had so many officers and their protocols were very well established, it would not be enough to regulate the vehicle flows properly. This research work is about something similar: the effective application of artificial intelligence methods to automate the control of Internet flows when the networks experience unforeseen situations. The proposed solutions allow the network administrators to manage some network tasks more efficiently, with minimal intervention, and focus on the situations where the human involvement is critical.
Gomez, Cesar A., "Leveraging Machine Learning Techniques towards Intelligent Networking Automation" (2021). Electronic Thesis and Dissertation Repository. 8091.
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 4.0 License.