Exploiting Heterogeneity in Modern Data Systems

Modern data systems increasingly operate in heterogeneous environments—spanning edge devices to cloud servers, leveraging diverse compute architectures, integrating multiple data platforms, and processing multi-modal data. Traditional approaches that assume homogeneity limit both system performance and user accessibility. 
In this talk, I will present work across two research threads that explore how to intelligently exploit different dimensions of heterogeneity to make data processing more efficient and user-friendly. First, I will introduce NEMO (VLDB 2024), a scalable approach for placing decomposable aggregation operators in geo-distributed stream processing systems. NEMO efficiently handles millions of heterogeneous nodes across edge-fog-cloud topologies, achieving up to 6x lower latency while preventing node overload. Second, I will discuss ongoing work in spatial analytics, which addresses heterogeneity across compute resources, platforms, and data modalities. Together, these efforts demonstrate that embracing heterogeneity in modern data systems can lead to significant improvements in performance and usability.

Bio:
Eleni Tzirita Zacharatou is a W2 Assistant Professor at the Hasso Plattner Institute (HPI) and the University of Potsdam, where she leads the Spatial Analytics and Large-Scale Data Processing research group. She has also been a Junior Fellow at the Berlin Institute for the Foundations of Learning and Data (BIFOLD) since 2021. Prior to joining HPI in 2025, she was an Assistant Professor at the IT University of Copenhagen (2022-2025) and a Postdoctoral Researcher at TU Berlin (2019-2022). She earned her PhD from EPFL in 2019, under the supervision of Prof. Anastasia Ailamaki, and holds a Diploma M. Eng. in Electrical and Computer Engineering from the National Technical University of Athens (NTUA). Her research focuses on improving the efficiency of data analysis and supporting data-driven decision-making, particularly for spatial data, by developing data management tools that are aware of resource constraints, system and hardware capabilities, and data modalities.