ABSTRACT
Physical science and engineering communities have used and driven high-performance computing (HPC) for many decades. PSC, whose scope spans a broad portfolio of research and services in scalable analytics, simulation, and modeling, integrating HPC, artificial intelligence (AI), and data to enable discovery, designed Bridges to meet the requirements of nontraditional HPC communities. Prior to Bridges, PSC began receiving many requests for persistent databases on Blacklight, PSC’s previous large-memory system. Data is foundational to Bridges for its role in driving analytics. All data is online, not in an archival system, to enable low-latency, high-bandwidth access. Science is driven by data, and AI is most effective when presented with substantial datasets on which to train. Virtualization serves two main roles on Bridges. Primarily, virtual machines are deployed on a persistent, project-specific basis to provide encapsulation, specific software environments, and security. A significant difference between traditional and nontraditional HPC communities is their difference in preferred programming models and languages.
