This thesis project takes as a use case an Industrial IoT setup developed and deﬁned by Ericsson Lab, which aims at providing a reliable generic orchestration system that is able to orchestrate IoT applications on edge devices within a cloud edge. The purpose is to research how to split the workload in the network edge via various computational technologies, such as machine learning and linear or non-linear optimization methods, and select and implement one that would allows faster and more energy eﬃcient computations within this speciﬁc network architecture. After a presentation of the problem in question and the possible solutions to overcome it, the following steps are performed to conduct the thesis project:
• Understanding and presenting how central and edge clouds work in general;
• Evaluating how automated service deployment works in cloud-based infrastructures;
o Investigating how to use orchestration on a specific scenario provided by Ericsson Lab.
• As a related example, experimenting the orchestration mechanism with a selected application in the cluster’s edge nodes;
• Designing, implementing and integrating a new service orchestrator engine with a custom control logic, that manages the deployment of the services to the edge devices. The scheduler should be able to consider real-time network descriptors, and also have a fall-back option to a predefined scheduling behaviour;
• Evaluating the implemented custom scheduler’s dynamic behaviour (e.g. high availability, latency, workload balancing) in a selected edge computing use case.