Topology-Aware Service Function Chain Scheduling using DRL

Abstract

Advancements in technology have led to increased demands on network infrastructure. To manage this flexibly and effectively, concepts such as Network Function Virtualization (NFV) and Service Function Chain (SFC) have emerged. However, without efficient operation, these can lead to high costs in terms of power consumption, service availability, Quality of Service (QoS). Previous studies have attempted to address these challenges using Deep Reinforcement Learning (DRL) and Graph Neural Networks (GNN). Nevertheless, they have not clearly defined the relationships between nodes in clusters and containers in services, nor proposed effective methods for embedding these relationships, resulting in limitations for efficient scheduling. Therefore, we propose a system that defines these relationships in the form of a graph and embeds them effectively. Moreover, previous research was limited to simulations during the validation phase and did not involve actual network traffic collection, which restricted the ability to guarantee direct performance. In this study, we built a system to apply this in a real environment and, through actual performance validation, were able to reduce latency and power consumption by 5% while maintaining meaningful service availability.