Cray ClusterStor

Cray ClusterStor. It is a high-performance parallel computing storage solution designed for exascale computing environments, particularly within supercomputing and high-performance computing (HPC). The system integrates the Lustre file system with purpose-built hardware and software to deliver extreme I/O bandwidth and capacity. Developed initially by Cray Inc. and continued by Hewlett Packard Enterprise following its acquisition, it became a foundational storage component for many leading TOP500 systems.

Overview

Cray ClusterStor was engineered to address the critical I/O bottleneck in modern supercomputing architectures, where compute node performance often outstripped traditional storage capabilities. It is fundamentally a scalable storage area network built around the open-source Lustre parallel file system, which is widely adopted in facilities like the National Energy Research Scientific Computing Center and the Texas Advanced Computing Center. The platform combines specialized server hardware, high-speed interconnect technology, and integrated management software to provide a unified namespace capable of supporting petabytes of data at hundreds of gigabytes per second. Its design philosophy emphasizes reliability and ease of management within complex data center environments, making it a preferred choice for large-scale scientific computing projects.

Architecture

The architecture of the system is modular, typically comprising object storage target (OST) servers, metadata server (MDS) units, and high-performance storage devices, all interconnected via a low-latency fabric. Early deployments utilized InfiniBand networks, while later generations integrated Cray Slingshot, the proprietary Ethernet-based interconnect developed by Cray Inc. for its Cray XC and Cray Shasta systems. The hardware employs dense JBOD enclosures populated with high-capacity hard disk drives and, increasingly, NVMe solid-state drives for tiered performance. This scale-out architecture allows incremental expansion of both capacity and bandwidth by adding additional storage modules and Lustre router nodes, ensuring linear performance scaling for applications running on systems like the Cori (supercomputer) at Lawrence Berkeley National Laboratory.

Software and Management

The software stack is centered on a robust implementation of the Lustre file system, managed through proprietary tools such as the Cray DataWarp software and the ClusterStor Management Console. These tools provide comprehensive monitoring, provisioning, and fault management capabilities, simplifying administration for sites like the Oak Ridge National Laboratory and the Argonne National Laboratory. Key software features include advanced data integrity field protection, integrated performance monitoring, and support for HPE's HPE Performance Cluster Manager. The system also supports various data protection and archive strategies, interfacing with HPSS and other tape library systems, which is critical for long-term data retention in projects funded by the United States Department of Energy and the National Science Foundation.

Deployment and Use Cases

Cray ClusterStor systems have been deployed in numerous flagship supercomputing centers worldwide. Major installations include storage for the Cray XC40-based Trinity (supercomputer) at Los Alamos National Laboratory and the Cray XC50 system at the Swiss National Supercomputing Centre. Its primary use cases involve data-intensive scientific simulations for fields like climate modeling, computational fluid dynamics, genomics, and high-energy physics, where sustained high-throughput I/O is paramount. The platform also supports emerging workloads in artificial intelligence and machine learning, providing the necessary data pipeline for training large models at institutions like the Pittsburgh Supercomputing Center.

History and Development

The product line originated from Cray's earlier Cray Sonexion storage system and was formally launched as Cray ClusterStor in 2012. Its development was closely aligned with the evolution of Cray's Cray XC series supercomputers and the Lustre community roadmap. Following the acquisition of Cray Inc. by Hewlett Packard Enterprise in 2019, the technology was integrated into HPE's broader HPC portfolio. Subsequent development focused on tighter integration with the HPE Cray EX platform and the Slingshot interconnect, culminating in storage solutions for exascale systems like the Frontier (supercomputer) at Oak Ridge National Laboratory. The lineage of ClusterStor technology continues within Hewlett Packard Enterprise's current high-performance storage offerings.

Category:Supercomputer storage Category:Parallel computing Category:Hewlett Packard Enterprise