Managing Event-Driven Applications in Heterogeneous Fog Infrastructures

The steady increase in digitalization propelled by the Internet of Things (IoT) has led to a deluge of generated data at unprecedented pace. Thereby, the promise to realize data-driven decision-making is a major innovation driver in a myriad of industries. Based on the widely used event processing paradigm, event-driven applications allow to analyze data in the form of event streams in order to extract relevant information in a timely manner. Most recently, graphical flow-based approaches in no-code event processing systems have been introduced to significantly lower technological entry barriers. This empowers non-technical citizen technologists to create event-driven applications comprised of multiple interconnected event-driven processing services. Still, today’s event-driven applications are focused on centralized cloud deployments that come with inevitable drawbacks, especially in the context of IoT scenarios that require fast results, are limited by the available bandwidth, or are bound by the regulations in terms of privacy and security. Despite recent advances in the area of fog computing which mitigate these shortcomings by extending the cloud and moving certain processing closer to the event source, these approaches are hardly established in existing systems. ... mehrInherent fog computing characteristics, especially the heterogeneity of resources alongside novel application management demands, particularly the aspects of geo-distribution and dynamic adaptation, pose challenges that are currently insufficiently addressed and hinder the transition to a next generation of no-code event processing systems.

The contributions of this thesis enable citizen technologists to manage event-driven applications in heterogeneous fog infrastructures along the application life cycle. Therefore, an approach for a holistic application management is proposed which abstracts citizen technologists from underlying technicalities. This allows to evolve present event processing systems and advances the democratization of event-driven application management in fog computing. Individual contributions of this thesis are summarized as follows:
1. A model, manifested in a geo-distributed system architecture, to semantically describe characteristics specific to node resources, event-driven applications and their management to blend application-centric and infrastructure-centric realms.
2. Concepts for geo-distributed deployment and operation of event-driven applications alongside strategies for flexible event stream management.
3. A methodology to support the evolution of event-driven applications including methods to dynamically reconfigure, migrate and offload individual event-driven processing services at run-time.

The contributions are introduced, applied and evaluated along two scenarios from the manufacturing and logistics domain.