High-performance event processing on modern hardware

PI: Bernhard Seeger (Uni Marburg)
Project Collaborator: Nikolaus Glombiewski
Project website: https://www.uni-marburg.de/en/fb12/research-groups/dbs/research/high-performance-event-processing-on-modern-hardware

We propose a system that deals with both low latency requirements of Complex Event Processing (CEP) and high-throughput analysis of historical (event) data in a reactive infrastructure monitoring scenario at the European Organization for Nuclear Research (CERN). Our core research objective is to combine both approaches through the introduction of the situation calculus into the world of CEP, through leveraging the presence of historical data for real time processing purposes and through the means of modern hardware. In order to improve low-latency of event processing, we develop novel pattern matching algorithms capable of dealing with situations using new co-processor technologies. In order to increase throughput, we substantially enhance our event store ChronicleDB into the following two directions. First, we develop new indexing capabilities by exploiting modern storage technologies like SSDs. In particular, our focus is on the design of new loading strategies for multiversion indexes and on the improvement of the layout of indexes on SSDs to exploit their inherent parallelism for processing queries like pattern matching. Second, we make ChronicleDB a scalable distributed system with new strategies for distributing event streams and indexes. Moreover, optimization strategies are developed that control the configuration and processing of queries in our system. We evaluate the performance and technical capabilities of our system based on a grand challenge in demanding infrastructure-monitoring applications at CERN.