Serverless edge swarm orchestration with federated reinforcement learning for communication networks

This paper proposes a vertical orchestration and reinforcement for task execution (VORTEX) framework, which is a fully distributed serverless edge swarm orchestration framework that integrates federated reinforcement learning (FRL) for adaptive function placement in heterogeneous communication networks, particularly targeting environments comprising diverse devices such as IoT gateways, edge servers, and smart cameras. To enhance clarity over traditional systems, unlike centralized schedulers that suffer from scalability and privacy limitations, VORTEX enables each edge device to locally learn a suitability scoring policy based on real-time hardware states, workload conditions, and function profiles. A global orchestrator aggregates these policies through a customized vertical federated learning (VFL) mechanism. By jointly optimizing quality-of-service (QoS) compliance and resource efficiency, the proposed method dynamically maps function invocation requests to the most appropriate devices under strict latency constraints. Extensive simulations demonstrate that VORTEX consistently outperforms baseline strategies in both reliability and energy efficiency. Specifically, the proposed approach achieves the lowest QoS violation rates across all tested parameter intervals, with violations decreasing from approximately 85% to 30% as system conditions improve, while competing methods remain significantly higher. Moreover, energy consumption analysis shows that VORTEX reduces median energy usage by up to 20–25% compared with conventional orchestration schemes, while also exhibiting lower variability and improved stability. These results confirm that FRL-driven decentralized orchestration effectively enhances scalability, robustness, and sustainability in dynamic edge computing environments.