MRI: Acquisition of the Kentucky Research Informatics Composable Cloud (KyRICC)

University Of Kentucky Research Foundation, Lexington KY

Investigators

Abstract

Scientific discovery today is driven by computation and data-intensive research that exploits the growing amounts of available data. However, the wide variety and size of emerging datasets often make analysis challenging on current high-performance computing (HPC) infrastructures because system configurations cannot be customized to process the data efficiently. This project will acquire and deploy a dynamically composable computer infrastructure called the Kentucky Research Informatics Composable Cloud (KyRICC). This KyRICC architecture will support complex data analysis pipelines with highly heterogeneous hardware requirements not currently supported by current HPC infrastructures. As a result, this project will enable and support a wide range of new research activities. KyRICC will be used by hundreds of University of Kentucky (UK) researchers (faculty, staff, and students) and by other computational research collaborators at institutions across the region including Centre College, Morehead State University, Eastern Kentucky University, University of Louisville, Northern Kentucky University, and Kentucky State University. As a leading-edge system, KyRICC and the exciting projects it makes possible will help recruit students, including students from groups underrepresented in STEM, to computational research. The system will also enhance the research training of many undergraduates, graduate students, and postdocs in Kentucky colleges and universities. The KyRICC architecture will support complex data analysis pipelines with highly heterogeneous hardware requirements across individual data analysis steps. Specifically, KyRICC will integrate four subsystems that will enable dynamically composable cloud infrastructure: (1) A cluster of peripheral-composable compute nodes, allowing for up to 10’s of GPUs and 10’s of TB of main memory on a single node. Groups of nodes can be dynamically allocated to allow the training and inference of very large deep learning models and datasets; (2) A next-generation high-speed NVMe-based storage cluster capable of efficiently serving large volumes of data to multi-GPU nodes. Unlike traditional clustered storage systems, this composable filesystem allows the partitioning of storage on the project-level, allowing us to isolate data and better manage system performance; (3) A Peta-scale storage system provided by UK’s current research storage infrastructure, providing a total of 2.2 PB of storage; and (4) An innovative workload management system for dynamic infrastructure composition, workload profiling, model and infrastructure tuning, supporting common pipelines and machine and deep learning models through templated projects. KyRICC will be a regional computational resource and will also be made available to the broader national computational research infrastructure through the NSF-supported ACCESS projects. Areas of expected breakthroughs in KyRICC-enabled research include deep learning and computer vision; natural language processing and multimodal embedding; computational modeling and simulation with data analytics; and omics analysis and systemic integration. This project is jointly funded by the Major Research Instrumentation (MRI) program, the Established Program to Stimulate Competitive Research (EPSCoR), and the Computer & Information Science & Engineering (CISE) Directorate. 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.