Desigining and validation of a flexible Reinforcement Learning agent for Network Slicing Architectures

• Grant number: 24/08973-2
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: July 01, 2024
• End date: June 30, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science
• Agreement: MCTI/MC
• Principal Investigator: Flávio de Oliveira Silva
• Grantee: Marcus Vinicius Diniz dos Reis
• Host Institution: Faculdade de Computação (FACOM). Universidade Federal de Uberlândia (UFU). Ministério da Educação (Brasil). Uberlândia , SP, Brazil
• Associated research grant: 18/23097-3 - SFI2: slicing future internet infrastructures, AP.TEM

Abstract

This Scientific Initiation (IC) work plan within the scope of the SFI2 project aims to build and validate Reinforcement Learning mechanisms for autonomously and intelligently orchestrating network slices. In the state of the art, approaches and constructions of intelligent mechanisms for orchestrating network slices can be found, however, they are inflexible in terms of the utilization and coexistence of different Artificial Intelligence techniques that the orchestrator may use. Flexibility in this context is fundamental because in each phase of network slice management there are heterogeneous demands that only the native coexistence of intelligent algorithms can support. Therefore, the main objective of this research is to advance the proposition of Reinforcement Learning algorithms that can be directly integrated into the SFI2 Orchestrator. The main challenge of the plan is to find good parameters for the Reinforcement Learning agent to deal with the action and reward space in a dynamic communication scenario while honoring service level agreements. Experimental evaluation will be conducted in a realistic scenario, with a communication network exhibiting behaviors close to a real network, specifically in the FIBRE-NG testbed of the National Research and Education Network (RNP). The first phase of the methodology of this research plan consists of studying and surveying the state of the art on techniques for guaranteeing and negotiating service level agreements in network slicing architectures followed by a study on isolation mechanisms and Artificial Intelligence applied to these architectures. In the second phase, algorithms based on Reinforcement Learning capable of adapting to the dynamic scenario of a production network to ensure service level agreements will be proposed and validated. In the third phase, integration and testing activities will occur on microservices architectures aiming at incorporating the solution into the SFi2 Orchestrator. At the end of this research, it is expected to achieve results that are complementary to the objectives of SFi2, as well as to provide human resources training. (AU)