Design and Implementation of a Cost-Effective Failover Mechanism for Containerized UPF

초록

Private 5G networks offer exclusive, secure wireless communication with full control deployments for many clients, such as enterprises and campuses. In these networks, edge computing plays a critical role by hosting both application services and the User Plane Functions (UPFs) as containerized workloads close to end devices, reducing latency and ensuring stringent Quality of Service (QoS). However, edge environments often face resource constraints and unpredictable failures such as network disruptions or hardware malfunctions, which can severely affect the reliability of the network. In addition, existing redundancy-based UPF resilience strategies, which maintain standby instances, incur substantial overheads and degrade resource efficiency and scalability for the applications. To address this issue, this study introduces a novel design that enables quick detection of UPF failures and two failover mechanisms to restore failed UPF instances either within the cluster hosting the failed UPF or across multiple clusters, depending on that cluster's resource availability and health. We implemented and evaluated our proposed approach on a Kubernetes-based testbed, and the results demonstrate that our approach reduces UPF redeployment time by up to 37% compared to baseline methods and lowers system cost by up to 50% under high-reliability requirements compared to traditional redundancy-based failover methods. These findings demonstrate that our design can serve as a complementary solution alongside traditional resilience strategies, offering a particularly cost-effective and resource-efficient alternative for edge computing and other constrained environments.

키워드