Towards AI-based Network Programmability as an enabler for Zero-touch management and orchestration in B5G infrastructures

Resumen

The emergence of next-generation networks (NGNs) presents significant challenges for communication infrastructures, needing increased flexibility and autonomous orchestration to achieve higher speeds and lower latencies. These networks will be crucial for futuristic services across various sectors, supported by Artificial Intelligence (AI) for cost-effective maintenance. The proliferation of virtualization technologies in these areas poses interoperability challenges, which AI-powered management and orchestration functions can address through automated decision-making. The evolution of AI-based network functions will enable predictive orchestration, optimizing traffic management, resource placement, and service configuration based on expected needs. This will lead to the realization of zero-touch management, minimizing human intervention and maximizing network efficiency. Furthermore, the convergence of traditional networking and computing architectures toward a unified computing continuum will democratize resource usage, increase response capacity, and reduce client-server traffic flows, thereby decreasing network latency. To advance the state of the art and address these challenges, the main objective of this thesis is to explore the current needs of NGNs to adopt fully functional Zero-touch network and Service Management (ZSM) capabilities. It is intended to provide the integration of different technologies that permit the fast management of the data plane, e.g., efficient packet processing, directed by AI-driven network functions enabling the autonomous management of the network in real-time. To reach those objectives, the performed work was divided into different research lines that converged to compose this doctoral thesis. In first place, there was a search and study phase, in which the literature was reviewed to deeply understand the state of the art of the ZSM concept, AI applied to networking, standardization activities, and ongoing research projects for NGNs. Once this was done, the tools and technologies most prominently used in these areas were explored and examined, with the aim of selecting the most proper ones to achieve intelligent network reprogrammability in ZSM scenarios. eBPF was the selected packet-handling technology after experimentally comparing it to P4. eBPF evidenced it is a flexible, simple and cost-effective solution that may be deployed in multiple scenarios without presenting prohibitive hardware requirements. By using eBPF as a traffic processing technology, the next stage was to select the proper Machine Learning (ML) algorithms to analyze the traffic and automatically react to changes, disruptions, or threats in the network. The initial literature review revealed that NNs are one of the fundamental pillars to enable smart decision-making in ZSM environments. In this way, a development was made to combine eBPF packet processing with neural networks to obtain ML-enabled networking capabilities. As a result, a packet handling pipeline based on ML was integrated within the Linux kernel, thus accelerating the efficient execution of AI algorithms. To validate the developed implementation, a real-world use case was explored, in which the solution examined real-time traffic while using a neural network to detect attacks. Finally, with all the knowledge acquired during the development of the research lines, a final exercise was made to envision the key networking and computing aspects that will drive the design of future 6G infrastructures. The stringent and challenging automotive vertical was studied from the point of view of its future applications and services to explore how they will shape the 6G network’s definition and deployments.