A Hybrid Auto-Scaling Method for Energy Efficiency of Network Services

In modern communication networks, supporting flexible and agile operations is crucial for meeting the diverse requirements of services in response to dynamic user demands and varying network conditions. Auto-scaling, which dynamically adjusts the resources allocated to network services, is a key enabler for addressing these issues. However, the auto-scaling problem has become more complicated due to additional considerations, including energy savings, energy efficiency, and carbon emissions, which arise from the need to address climate change. To tackle these challenges, this paper proposes a hybrid auto-scaling approach that combines reinforcement learning with a greedy algorithm. The proposed approach dynamically adjusts the number and placement of instances constituting a network service, thereby reducing energy consumption and carbon emissions while meeting service performance requirements.