Efficient Resource-Aware Proactive Flow Rule Caching in Software-Defined Access Networks
Youngjun Kim, Tae-Kook Kim, Yeunwoong Kyung · 2025
As Software-Defined Networking (SDN) becomes increasingly critical for managing complex and dynamic network environments, efficient flow rule caching has emerged as a key challenge. Traditional reactive caching approaches introduce latency during handovers, impacting the Quality of Service (QoS) for delay-sensitive applications. Proactive caching methods address this by predicting mobile nodes’ (MNs) future locations; however, they often face issues related to prediction accuracy and memory utilization. In this paper, we propose a efficient resource-aware proactive flow rule caching based on multi-agent reinforcement learning (MARL). Our approach dynamically predicts MN movement patterns, enabling the SDN controller to preinstall flow rules in a targeted and timely manner. A hierarchical multi-agent architecture is introduced to adjust caching strategies based on the mobility level of MNs, maximizing flow setup hit ratio (FSHR) while minimizing unnecessary TCAM occupancy. Proposed MARL-based caching strategy presents a scalable and efficient solution for flow management in SDN, particularly in environments requiring high mobility and low latency.