BIFOLD Colloquium 13/2025

Extreme-Scale Time Series Management and Analytics with ModelarDB: A Tale from Denmark

Abstract:

Modern wind turbines are monitored by sensors which generate massive high-frequency time series that are ingested on edge devices and then transferred to the cloud where they are stored and used for analytics, e.g., ML models, for decades. Current systems have four challenges: (1) limited edge hardware struggles to keep up with ingestion; (2) limited bandwidth means not all data can be transferred; (3) high storage costs on both edge and cloud; and (4) low data quality for analytics. This talk will introduce the Time Series Management System ModelarDB, developed at Aalborg University, which handles all four challenges by efficiently managing time series across the entire pipeline. The talk will explain the main ideas of ModelarDB, its architecture and open source implementation over two generations, and how it achieves a sweet spot of high compression, fast ingestion, and fast analytics queries. The talk will also cover how ModelarDB’s error-bounded lossy compression interacts with time series forecasting using ML models and how efficiency and analytics accuracy can be co-optimized.

Speaker

Torben Bach Pedersen is a Professor in the Data Engineering, Science, and Systems Group at the Department of Computer Science,  Aalborg University. His research interests include predictive, prescriptive, and extreme-scale data analytics with digital energy as the main application area. He is an ACM Distinguished Scientist, an IEEE Computer Society Distinguished Contributor, an AAIA Fellow, a Member of the Danish Academy of Technical Sciences, and received an Honorary Doctorate from TU Dresden for this contributions to data analytics and digital energy. He is currently on a sabbatical at the DAMS Group at BIFOLD.