Collaborative research: elements: computational storage virtualization for accelerating data-driven scientific applications on supercomputers

The University of Houston System received a $120,621 Project Grant award from the National Science Foundation's Computer and Information Science and Engineering program (CFDA 47.070) on June 15, 2025, with a completion date of May 31, 2028. This collaborative research initiative develops ?-HDF5, a virtual computational storage system designed to optimize data-driven scientific applications on supercomputers by integrating computational storage devices with high-performance computing (HPC) infrastructure. The project delivers three primary products: an HDF5-compatible interface supporting multiple compute kernels, a kernel dispatch and execution system for heterogeneous I/O layers in HPC systems, and efficient management solutions for distributed and heterogeneous storage devices handling both raw and pre-processed data.

The research outputs address critical infrastructure needs for scientific computing across multiple domains including national security, power system reliability, and food security. By making computational storage devices more accessible to scientific computing communities, ?-HDF5 enables accelerated data ingestion pipelines and improved system-level support for data-driven applications. The project also provides educational components, including research training in Very Large Scale Integration (VLSI) and artificial intelligence (AI) for undergraduate students, establishing educational pathways toward advanced degree programs in high-performance computing and systems research.

Abstract

With the emergence of computational storage devices and smart storage solutions, new system-level support is needed to enable data-driven scientific applications to efficiently access and utilize underlying compute and storage resources. This project contributes by creating a new basic toolchain and system technologies for such systems and by evaluating them with representative scientific applications and data analytics. IT has the potential to impact other scientific domains because data-driven scientific applications are used in many scientific and engineering domains, including national security, power system reliability, and food security. The project will provide research training in very large scale integration (VLSI) and AI for undergraduate students and provide them with an educational pathway to pursue advanced degrees. This collaborative and interdisciplinary project seeks to bring experts in computer systems, field-programmable gate arrays (FPGAS), high-performance computing (HPC), and domain scientists together to design and implement a virtual computational storage system ?-HDF5. The project is driven by real use cases and built on state-of-the-art HPC and machine-learning cyberinfrastructure. The project has three specific goals. First, the project will develop an HDF5-compatible interface and provide support for a wide variety of computer kernels. Second, IT will identify, dispatch, and execute compute kernels on faster devices across multiple I/O layers in HPC systems. Third, IT will efficiently manage CO-located raw data and pre-processed data on distributed and heterogeneous storage devices. Scientific applications using ?-HDF5 for data management can benefit from accelerated data ingestion pipelines. ?-HDF5 also makes computational storage devices more accessible to users in scientific computing communities. 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.- subawards are not planned for this award.