Systems

Cray partners with leading storage hardware and software vendors in order to create HPC storage solutions that balance performance with reliability and ease-of-management. From high-performance disk storage to tape storage for backup and archival needs, Cray can customize a solution which meets the workload and operational requirements of your HPC platform.

Local (Scratch) Storage: Local or scratch storage is required for data that needs to be directly accessible to the computational platform. It is generally used to store intermediate results from large calculations and application restart dumps. It is critical that I/O for this storage be very high performance or it will slow the application down. Scratch storage is generally implemented through on-node or Direct-Attached Storage solutions and dedicated to the HPC system — not shared with other platforms on the network.

Site-wide Shared Storage: Shared storage is the foundation of a data center, as it holds data accessed by all platforms (and all operating systems) on the network, including home directories and persistent data. It stores frequently-used files, permanent files, application input files, and the results of application runs. Site-wide shared storage is generally implemented through network-attached storage (NAS), or storage area networks (SAN).

Data Migration/Archive Storage: Archival storage is used for permanent data, including data and information produced by supercomputer runs, data gathered from instruments, legacy software, and data that doesn't change or is infrequently used. Like site-wide shared storage, this data must be accessible by all platforms on the network.