Towards Monitoring-Driven and Workload-Aware Scheduling for Federated Learning in Cloud-Edge Environments

The increasing adoption of Federated Learning (FL) in cloud and edge environments has intensified the need for deployment frameworks that can support dynamic, heterogeneous, and performance-sensitive execution scenarios. While several FL frameworks provide rich abstractions for algorithm design and simulation, the orchestration of real-world FL workloads still relies on generic cluster schedulers that are largely unaware of the communication patterns, synchronization phases, and resource contention characteristics of federated workflows. In this paper, we present an integrated solution that combines FLeeT, a modular and container-based framework for the automated deployment of FL pipelines, with a dedicated, monitoringdriven scheduling layer specifically designed for FL workloads. The proposed scheduler extends standard Kubernetes-based orchestration with intelligent placement strategies that jointly account for resource contention and communication locality, aiming at minimizing the end-to-end execution time of distributed federated workflows. The resulting system enables adaptive and performance-aware deployment of FL applications across heterogeneous cluster infrastructures, and represents a practical step towards computational-intelligence-driven resource management for large-scale and dynamic distributed learning systems, in line with the objectives of Cloud and Edge environments.