This repository houses QONNECT-FL, an extension of the QONNECT[1] orchestrator designed for the public cloud. It includes complete setups for two orchestration profiles—each spanning a cloud, edge, and fog cluster on Microsoft Azure—along with the required architectural adaptations for inter-cluster consistency.
It integrates a federated forecasting approach for resource-aware scheduling, alongside benchmarks for five deep forecasting architectures for cluster CPU load using heterogeneous public telemetry.
Detailed usage instructions can be found in the respective subdirectories:
-
deploy-QONNECT-fl/— Contains the infrastructure provisioning, Kubernetes cluster deployment, and the rollout of the QONNECT components. -
runtime-optimization/— Implements various deep time-series models trained via the tsai[2] framework (both standalone and federated using Flower[3]), serving as the predictive component for the resource orchestration approach in QONNECT.