HPE Cray OS

HPE Cray OS is a purpose-built, Unix-like operating system designed for HPE Cray Supercomputer systems. It is engineered to deliver extreme performance, scalability, and reliability for High-performance computing workloads in fields like Climate modeling, Computational fluid dynamics, and Genomics. The OS represents the culmination of decades of Cray Inc. system software expertise, now under the stewardship of Hewlett Packard Enterprise following its acquisition.

Overview

HPE Cray OS serves as the foundational software layer for modern Cray EX and Cray XE supercomputing platforms, managing the complex interplay between Manycore processors, high-speed interconnects like Slingshot, and massive-scale Parallel computing applications. It is optimized for Massively parallel architectures, providing a stable and efficient environment for running Message Passing Interface and Partitioned global address space programming models. The system is deployed at major national laboratories and research institutions worldwide, supporting mission-critical scientific discovery.

Architecture and Design

The architecture is a streamlined, Monolithic kernel design that minimizes Operating system overhead to maximize computational throughput for floating-point-intensive workloads. It integrates deeply with Cray Linux Environment components and the HPE Cray Programming Environment to provide a cohesive software stack. Key design tenets include robust Fault tolerance mechanisms, advanced Job scheduler capabilities via Slurm Workload Manager, and optimized support for heterogeneous compute elements including AMD EPYC processors and NVIDIA Grace Hopper GPUs. The design prioritizes direct, low-latency access to system resources over general-purpose features.

Key Features and Components

Notable features include the Cray DataWarp burst buffer technology for accelerated I/O, the Cray ATP abnormal termination processing tool for debugging large-scale jobs, and the Cray PMI library for efficient Message Passing Interface launch and management. The Craypat performance analysis suite and the Cray Reveal tool for directive-based optimization are integral components. The system also incorporates the Cray Urika graph analytics software suite and supports containerized workloads through integration with Singularity and Charliecloud.

Deployment and Usage

HPE Cray OS is deployed on some of the world's most powerful supercomputers, including systems at the Argonne National Laboratory, Oak Ridge National Laboratory, and the Lawrence Livermore National Laboratory. It is used for grand challenge simulations in Astrophysics, Materials science, and Energy research, such as those conducted for the United States Department of Energy's Exascale Computing Project. The National Oceanic and Atmospheric Administration utilizes it for weather forecasting, while institutions like the Pittsburgh Supercomputing Center support diverse academic research.

Historical Development and Versions

Its lineage traces directly to the UNICOS operating system developed by Seymour Cray's original Cray Research, evolving through versions for the Cray T3E, Cray XT, and Cray XC series. A major consolidation occurred with the introduction of the Cray Linux Environment, which unified system management. The rebranding to HPE Cray OS followed the acquisition of Cray Inc. by Hewlett Packard Enterprise in 2019. Subsequent development has focused on exascale readiness, with enhancements for the Frontier and Aurora systems.

Relationship to Other Cray Software

HPE Cray OS exists within a broader ecosystem of HPE Cray software. It is the core platform upon which the HPE Cray Programming Environment, including compilers like Cray Compiling Environment and libraries such as LibSci, is deployed. It is distinct from, but can interoperate with, the HPE Cray ClusterStor storage management software and the HPE Cray System Management tools. While it provides the base operating system, higher-level workload management and user environment are often facilitated through the Cray Linux Environment and interfaces like the NCSA Blue Waters portal.