Advancing HPC and ML Systems and Applications via Efficient Data Management

ABSTRACT
The new generation of supercomputers comprises exascale (10^18 floating-point operations per second) computer systems. These systems are instrumental in enabling scientists and engineers to tackle extremely complex high-performance computing (HPC) and machine learning (ML) problems, addressing critical societal challenges such as climate change, water management, advanced manufacturing, and vaccine and drug design. However, the gap between the ever-increasing compute power and the limited memory/storage capacity and I/O bandwidth necessitates the creation of intelligent and effective methods for efficiently managing the massive amounts of data generated by HPC and ML applications, ensuring fast storage and transmission. This talk will introduce our promising solution – error-bounded lossy compression – which significantly reduces data sizes while maintaining high data fidelity. This approach can greatly benefit data management, including I/O, memory, and storage, in many HPC and ML applications. The talk will cover the design, optimization, and application of our error-bounded lossy compression to advance HPC and ML systems, particularly in large-scale data processing applications such as HPC simulations, ML model training, and large graph analytics.

BIOGRAPHY
Dingwen Tao is an associate professor at Indiana University Bloomington, where he directs the High-Performance Data Analytics and Computing Lab. He received his Ph.D. in Computer Science from University of California, Riverside in 2018 and B.S. in Mathematics from University of Science and Technology of China in 2013. He is the recipient of various awards including NSF CAREER Award (2023), Amazon Research Award (2022), Meta Research Award (2022), R&D100 Awards Winner (2021), IEEE Computer Society TCHPC Early Career Researchers Award for Excellence in High Performance Computing (2020), NSF CRII Award (2020), and IEEE CLUSTER Best Paper Award (2018). He is serving as an Associate Editor of IEEE Transactions on Parallel and Distributed Systems. He was also a program committee member of major HPC venues such as SC, HPDC, ICS, IPDPS, etc. He is a senior member of IEEE and ACM.