Automated Pricing-based Provisioning of SDN/NFV Services in Distributed Multi-access Edge Computing using Cooperative Multi-Agent Deep Reinforcement Learning

The disruption caused by Software Defined Network and Network Function Virtualization (SDN/NFV) technologies will have many impacts on the telecom network. Specifically, the network architecture based on ETSI MANO comprising Virtual Infrastructure Manager (VIM), Virtual Network Function Manager (VNFM), and NFV Orchestrator (NFVO) will significantly change how we operate and manage the telecom network. These impacts on the network architecture will be made gradually. Thus, the migration from non-virtualized networks to all virtualized networks will happen step by step. By introducing new actors into the telecom ecosystem, NFV-MANO will bring in new business models. It is envisaged that these new actors/models will promote competition hence the demand for more flexible charging models with real-time charging. In this research, we will address the architectural realization of the SDN/NFV charging model under the business model of MANO (MANagement and Orchestration) when the service provider (SP) uses the Network Function Virtualization Infrastructure (NFVI) from the NFVIaaS Provider in a distributed multi-access edge computing (MEC) environment. To optimize the Quality of Service (QoS) of MEC resource allocation for incoming services and maximize the overall operating profit of the service provider adopting our charging model, we also propose a new cooperative multi-agent actor-critic based deep reinforcement learning (MADRL) method trained with proximal policy optimization algorithm, namely Coop MAPPO. The results of our experiments showcase the superiority of the Coop-MAPPO multi-agent system over alternative decision-making approaches, with its potential for enhancing operational efficiency and profitability while minimizing failure rates.