Collaborative Research: Elements: SGCC: An Efficient GPU-oriented Data Reduction Cyberinfrastructure for Scientific Data Analysis

The rapid development of GPU hardware has promoted scientific supercomputing, enabling exascale data production on heterogeneous supercomputing systems. With GPU dominance in heterogeneous computing, the cyberinfrastructure of GPU-based scientific data compressors is still maturing, and several gaps need to be addressed: existing frameworks lack adaptations to many scientific data analysis requirements; there are no user-friendly interfaces and off-the-shelf solutions for GPU-based scientific data compressors; and the compressors that support non-NVIDIA GPU architectures are very limited. This project develops a user-friendly, high-performance, and portable GPU-accelerated data reduction cyberinfrastructure for all primary GPU-equipped supercomputing systems. It will mitigate data challenges on GPU-equipped supercomputing systems, improve data analysis efficiency, and eventually accelerate scientific discovery. This project will continuously contribute to the education and training of graduate students by enhancing the quality of computing-related curricula in heterogeneous scientific computing, data management, and visualization. This project builds Scientific GPU Compression Cyberinfrastructure (SGCC), a user-friendly end-to-end cyberinfrastructure of GPU-based data compression for scientific data workflows, by porting, extending, and optimizing multiple existing capabilities, including but not limited to: the cuSZ family of error-bounded lossy compressors, GPU-based lossless encoders, QCAT (a CPU-based compression quality assessment toolkit), the Kokkos ecosystem (a multi-backend performance-portability framework), LibPressio (the unified programming interface of scientific compressors), and HDF5. To create SGCC, the project combines three thrusts: (1) SGCC ensures its efficiency and effectiveness in practical scientific data analysis workflows, providing adequate support for diverse data formats and compression quality targets; (2) SGCC improves the usability of the GPU-accelerated data-reduction ecosystem by providing high-level language bindings, command line interface, and user-interface integrated with visualization functionality; and (3) SGCC enables state-of-the-art GPU-accelerated scientific data compressors on multiple heterogeneous computing platforms, such as NVIDIA, AMD, and Intel. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.