Distributed scheduling in Kubernetes based on MAS for Fog-in-the-loop applications

With cloud computing gaining momentum in industrial environments, the next step seems to move the computing infrastructure closer to the devices that collect the data at plant level. Fog computing is a paradigm that takes advantage of the fast response times of working relatively close to the plant, and the storage, processing and availability features of the cloud. Fog computing can be used to improve the controllability of automation processes by introducing a higher-level control loop: Fog-in-the-loop (FIL). FIL allows capturing data from the plant, processing it to extract information and feedback actions to the plant based on the processing results. Therefore, FIL applications are context-aware applications that require the deployment of distributed components and dynamic reconfiguration. In this context, we are involved in a project that aims at integrating a model-based, multipurpose Multi-Agent System (MAS) platform with a containerized application orchestrator to meet the requirements of Fog-in-the-loop applications. Specifically, this paper describes a custom scheduler for Kubernetes orchestrator that distributes the scheduling task among the processing nodes by means of the MAS. This new scheduling approach proved to be faster than the centralized scheduling approach used by the default scheduler of K8s.