Federated Autonomous Orchestration in Fog Computing Systems

Autonomous computing is the key concept for service orchestration in next-generation cloud computing environments. Virtualization supplements it by adding an abstraction layer to the services, as well as to the underlying infrastructure including the network layer. IP flows are calibrated by Software Defined Networks to address virtualized services directly, which is conceptually described as Network Function Virtualization. Side by side exists a diversity of devices in the cloud-to-fog continuum, which are geographically distributed, can connect to a multitude of networks, and are prone to churn. Their behavior increases the manageability and maintenance efforts needed to deploy services dynamically. To achieve a MAPE-K like service distribution, we adopt machine learning concepts to react to infrastructure changes and service requirements. Due to the distributed nature, we focus on federated machine learning to improve overall performance without cluttering single nodes.