Seminar:
Designing Efficient Data Reduction Approaches for Multi-Resolution Simulations on HPC Systems

ABSTRACT

As supercomputers advance towards exascale, computational intensity increases significantly, and the volume of data requiring storage and transmission experiences exponential growth. Multi-resolution methods such as Adaptive Mesh Refinement (AMR) can effectively address these challenges. Concurrently, error-bounded lossy compression is one of the most efficient approaches to tackle the data volume issue. Despite their respective advantages, few studies have examined how the multi-resolution method and error-bounded lossy compression can work together. To bridge this gap, this study introduces a series of application-driven system solutions, coupled with algorithmic innovations, for real-world multi-resolution data reduction: (1) This study first enhances the offline compression quality of multi-resolution data for different state-of-the-art scientific compressors by adaptively preprocessing the data based on data features and optimizing the compressor. (2) This study then presents a novel in-situ lossy compression framework, utilizing HDF5 and enhanced SZ2, specifically tailored for real-world AMR applications. This framework can improve I/O costs and compression quality. (3) Finally, this study introduces a workflow for multi-resolution data compression applicable to both uniform and AMR simulations. It optimizes three compressors to better integrate multi-resolution techniques with lossy compression and incorporates an advanced uncertainty visualization method to help users understand the potential impacts of compression.

Daoce Wang

Indiana University

Daoce Wang is a fifth-year Ph.D. student at Indiana University, Bloomington. He earned his bachelor's degree in Computer Science from the University of Electronic Science and Technology of China in 2018 and his master’s degree in Computer Science from the University of Florida in 2020. He served as a summer research intern at Los Alamos National Laboratory in 2021, 2022, 2023, and 2024. His research interests include high-performance computing (HPC), scientific data management and visualization, lossy compression, machine learning (ML), and fault tolerance. Daoce’s primary Ph.D. research focuses on designing efficient data reduction methods for extreme-scale scientific simulations on HPC systems. He has published five first-authored papers in CLUSTER ’21, HPDC ’22, TPDS ’24, SC ’23, and SC ’24.