Computer Science Seminar by Tian Jiannan: Scientific Data Reduction in the Era of Exascale Computing
Speaker:
Title: Scientific Data Reduction in the Era of Exascale Computing
Abstract
Today鈥檚 scientific exploration is driven by large-scale simulations and advanced instruments, generating vast amounts of data at extreme rates. As we advance into the era of exascale computing (1e18 FLOPS), the data volume is projected to increase exponentially, imposing an unprecedented burden on supercomputing systems and becoming a bottleneck in scientific applications. At the same time, supercomputers and HPC applications are evolving to be more heterogeneous, incorporating accelerator-based architectures (e.g., GPUs), presenting significant complexities in managing data across the systems. With the lagging development of memory and storage systems compared to the compute capabilities, scientific data processing has emerged as an effective solution to address the scalability bottlenecks. This talk explores how a software-hardware co-design approach enables R&D in accelerated error-controllable lossy compression for scientific data with a balance of speed, fidelity, and compression ratio. Additionally, it also discusses the transformative potential for emerging technologies to deepen techniques in data processing and post hoc analytics, shedding light on the new research on the horizon.
Bio
Jiannan Tian is currently a postdoctoral appointee in the Department of Computer Science at the University of Kentucky (UKy), working with Dr. Xin Liang in scientific data reduction and analytics. Before joining UKy, he earned his Ph.D. from Indiana University Bloomington (IUB). He was also a long-term intern at Argonne National Laboratory, where he was working closely with scientists across multiple DOE (the Department of Energy) projects under the Exascale Computing Project. His research focuses on leveraging accelerators, particularly GPUs, to tackle data explosion challenges in scientific applications. He has authored over 30 publications in premier venues, including PPoPP, PACT, CLUSTER, IPDPS, HPDC, and SC. As the primary developer of the pSZ/cuSZ software, he advances error-controllable scientific lossy compression with acceleration. He also contributes to the SZ ecosystem for scientific compression and data analytics, which received the 2021 R&D 100 Award and the 2023 Best Paper Award for its emerging significance.