Energy Efficient VNF-FG Embedding via Attention-based Deep Reinforcement Learning

Designing smart mechanisms to facilitate and accelerate service deployment and management is one of the most challenging aspects for network infrastructure providers. This is due to the massive amount of traffic that they are expected to support, the decentralized nature of the architectures, and the services they run to meet quality targets and avoid Service Level Agreement (SLA) violations. Therefore, Communications Service Providers (CSPs) are devoting much of their efforts on reducing energy consumption and reducing carbon footprint of their network infrastructures. In future communication networks, traditional management mechanisms, and centralized legacy solutions show their limitations in ensuring revenue for the infrastructure providers, the service providers, and a good Quality of Experience (QoE) for the end-users. The deployment of these services requires, typically, an efficient allocation of Virtual Network Function Forwarding Graph (VNF-FG). In this context, we propose an intelligent energy efficient VNF-FG embedding approach based on multi-agent attention-based Deep Reinforcement Learning (DRL). Our contribution uses a semi-distributed DRL mechanism for VNF-FG placement. The proposed algorithm is shown to outperform previous state-of-the-art approaches in terms of acceptance rate, power consumption, and execution time.