Energy-Efficient Event Processing on Modern Hardware (SE553/9-2)
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
Complex Event Processing (CEP) is a widely adopted technology for low latency analysis of continuous data streams. It is a crucial component for many modern data processing use cases, including technical infrastructure monitoring, continuous log analysis and traffic monitoring. In previous projects, we developed a hybrid event processing and storage system for modern hardware. We were able to increase the throughput of continuous queries with minimal impact on result latency by utilizing integrated GPUs for query processing. Furthermore, the storage system leverages tailor-made index structures for long-term storage and brings efficient CEP queries into the database world.
Although those past efforts focused on efficient processing, an often overlooked aspect is energy efficiency. With growing concerns about IT’s carbon footprint and a general desire to save cost, we believe this will be an essential aspect in future research across multiple domains. Moreover, a growing heterogeneity of compute devices has lead to a wide spectrum of modern hardware configurations, ranging from efficient high-end GPUs to low-powered mobile devices. We want to explore how both ends of this spectrum can contribute to a reduced energy consumption in CEP use cases. For this purpose, we aim for the following three goals during this project. First, we will analyze and optimize our current hybrid event processing system with regards to energy consumption. Second, we want to develop energy-conscious CEP algorithms for both dedicated and integrated GPUs. Third, based on flight monitoring data, we will explore novel spatio-temporal use cases to improve and benchmark our developed solutions. The expected result of this project is an energy-conscious hybrid event processing system that can efficiently utilize modern hardware.