Efficient Software Architectures for Distributed Event Processing Systems

Director:Philippsen, M.
Period:November 15, 2010 - December 31, 2015
Coworker:Mutschler, C.

Localization Systems (also known as Realtime Location Systems, or RTLS) become more and more popular in industry sectors such as logistics, automation, and many more. These systems provide valuable information about whereabouts of objects at runtime. Therefore, processes can be traced, analyzed, and optimized. Besides the research activities at the core of localization systems (like resilience and interference-free location technologies or methods for highly accurate positioning), algorithms and techniques emerge that identify meaningful information for further processing steps. Our research focuses on automatic configuration methods for RTLSs as well as on the generation of dynamic motion models and techniques for event processing on position streams at runtime.
In 2011, we investigated whether events can be predicted after analyzing and learning event streams from the localization system at runtime. As a result, we are able to deduce models that represent the information buried in the event stream to predict future events.
We developed several methods and techniques in 2012 that process and detect events with low latency. Events (composite, complex) can be detected by means of a hierarchical aggregation of sub-events that themselves are detected by (several) event detectors processing sub-information in the event stream. This greatly reduces the complexity of the detection components and renders them fully maintainable. They even can use parallel or distributed cluster architectures more efficiently so that important events can be detected within a few milliseconds.
In 2013 we further minimized detection latency in distributed event-based systems: first, a new migration technique modifies and optimizes the allocation of software components in a networked environment at runtime to minimize networking overhead and detection latencies. Second, a speculative event processing technique uses conservative buffering techniques to exploit available system resources. We also created and published a representative data set (consisting of realtime position data and event streams) and a corresponding task description.
In 2014 we investigated fundamental approaches zu handle uncertainties (both w.r.t. the definition of event detectors and to the events). We implemented a promising prototype of an event-based system that is no longer deterministic but instead evaluates several possible system states in parallel to achieve a detection with a much higher robustness and correctness. The domain expert can parameterize the event detectors by attaching probabilities or probability functions to the generated events.
In 2015 we improved, optimized and published our approach. Furthermore we started to investigate approaches to learn optimal parameter sets for the event detectors. Thus, a manual adjustment and tuning of parameters (like thresholds) becomes unnecessary.
The project is a contribution of the Programming Systems Group to the IZ ESI

ESI Anwendungszentrum
watermark seal