CC* Data Storage: Institutional Storage for the University of Notre Dame (NDStore)

点击翻译按钮获取中文摘要

This project equips the Center for Research Computing at the University of Notre Dame (ND CRC) and its scientific users across all Notre Dame colleges and departments to enable transformative research in social and physical sciences and engineering domains through the acquisition of institutional storage called NDStore. Major beneficiaries of NDStore are researchers utilizing various research cores at Notre Dame, such as the Genomics and Bioinformatics Core and the Notre Dame Integrated Imaging Facility, as well as other researchers from University Centers and Institutes, such as the Institute for Data and Society, in addition to the broader national community via the Open Science Grid. Together, these different facilities and researchers generate hundreds of terabytes of data per year, and they enable expert users to address the most complex research problems of today’s world. The major capabilities provided by NDStore accelerate existing research otherwise throttled by insufficient storage capability. They also enable full data lifecycle at previously inaccessible scales, enable new national data-intensive collaborations, and incubate new research projects. NDStore brings to Notre Dame an additional 2 petabytes of storage capacity for data manipulation, curation, and long-term preservation, as well as 250 terabytes of fast scratch storage for machine learning-related workloads. NDStore is a highly available solution based on an open-source Ceph-based storage clustering standard. It was designed with the flexibility to meet the needs of researchers at various stages of their research. NDStore provides a clear benefit to researchers who generate data with various instruments in core facilities and need to transfer the data to their home directories for analysis and curation before the data is shared with their communities. Before this project was funded, the amount of storage provided by ND CRC to each faculty lab was not satisfactory for most of the users dealing with large data coming from instruments at core facilities, such as microscopes, sequencing machines, or other benchtop devices. In addition, ND CRC’s high-performance scratch storage system has been shared between high-performance computing and machine learning workloads; very often, mixing these workloads led to performance bottlenecks, negatively impacting all of the storage system users at Notre Dame. NDStore helps Notre Dame create an independent scratch system for machine learning workloads.Another important aspect of the intellectual merit of this project is the opportunity for the CRC to deploy NDStore in such a way that the entire data lifecycle at Notre Dame’s research enterprise is supported. Research data use cases at ND are highly diverse, complex, and heterogeneous. They differ in types of data captured, scientific instruments used, data processing and analyses conducted, policies and methods for data sharing and use, and, internal to the lab, cyberinfrastructure-related knowledge. Data life cycle stages include: 1) data capture; 2) initial processing near the instrument(s); 3) central processing at data centers or clouds; 4) data storage, curation, and archiving; and 5) data access, dissemination, and visualization. Until NDStore was deployed, Notre Dame infrastructure could adequately support only stages 1-3 and 5, with very minimal support for stage 4. NDStore fills this gap. NDStore will also be integrated into classroom and undergraduate internship programs hosted by numerous faculty in colleges and ND CRC. Through user training, research experience for undergraduates, pre-college programs for high school students, workshops, internships, and experiential training programs, ND CRC will ensure that NDStore has the broadest possible impact on the local and national academic community.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.