Design and Implementation of a Middleware for Uniform, Federated and Dynamic Event Processing

In recent years, real-time processing of massive event streams has become an important topic in the area of data analytics. It will become even more important in the future due to cheap sensors, a growing amount of devices and their ubiquitous inter-connection also known as the Internet of Things (IoT). Academia, industry and the open source community have developed several event processing (EP) systems that allow users to define, manage and execute continuous queries over event streams. They achieve a significantly better performance than the traditional store-then-process'' approach in which events are first stored and indexed in a database. Because EP systems have different roots and because of the lack of standardization, the system landscape became highly heterogenous. Today's EP systems differ in APIs, execution behaviors and query languages. This thesis presents the design and implementation of a novel middleware that abstracts from different EP systems and provides a uniform API, execution behavior and query language to users and developers. As a consequence, the presented middleware overcomes the problem of vendor lock-in and different EP systems are enabled to cooperate with each other. In practice, event streams differ dramatically in volume and velocity. We show therefore how the middleware can connect to not only different EP systems, but also database systems and a native implementation. Emerging applications such as the IoT raise novel challenges and require EP to be more dynamic. We present extensions to the middleware that enable self-adaptivity which is needed in context-sensitive applications and those that deal with constantly varying sets of event producers and consumers. Lastly, we extend the middleware to fully support the processing of events containing spatial data and to be able to run distributed in the form of a federation of heterogenous EP systems