Dynamic Network Slicing Using Deep Reinforcement Learning

• Autor(a) principal: Oliveira, Pedro Miguel Silvério de
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/155750
• topic: Network slicing; Deep Reinforcement Learning; Admission Control; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

Nowadays network slicing is one of the biggest drivers of new elements in the 5G network business. This is because this paradigm allows the creation of independent slices, with their virtually and logically separated radio, network and computational resources. Using network slicing, operators sell infrastructure resources of any kind to tenants, while tenants use these resources to sell services to their customers, the end users. In this context, a problem that is essential to solve is how to improve the operator’s profit, ensuring compliance with the requests’ SLAs and distributing network resources in order to increase its usage rate. This dissertation proposes to design two algorithms based on DRL for slice admission in the transport network, learning which request to accept and reject while guaranteeing the requirements of the tenants requests. The contributions of this study start with the formalization of the problem of slice admission, followed by its simulation and implementation of DRL agents using Containernet, the Ryu controller, OpenAI Gym and the PyTorch framework. The result is two DRL-based algorithms capable of achieving good performances in this simulated scenario.