June 2, 2014

HPC Storage Enters a New Era

Until relatively recently, HPC storage systems have been almost an afterthought, a grab bag mix of components jury rigged together to support the star of the show – a supercomputer or a large compute cluster.

A typical legacy HPC storage solution was made up of commercially available RAID arrays, network filers, or direct attached disks. The problem is that these were simply cobbled together to cover a point design in the past. However, these configurations are not able to efficiently deliver parallel I/O to thousands of compute nodes that we now commonly see in mid to large HPC system installations. The “I/O bottleneck” – the difficulty of moving reads and writes past the storage I/O interface controller – became a major sticking point. By themselves, these ad hoc separate storage piece-parts did not inherently scale to achieve desired characteristics such as massive linear performance scale from terabytes to petabytes storage capacity, along with a single globally coherent name space.

Not Quite There Yet

The development of Lustre® established the means to solve these problems – the Lustre file system has the right architecture to deliver the requisite massive performance and capacity needed, as well as the ability to present a single, globally coherent name space. Where things go wrong is in the implementation – continued use of the ad hoc approach described above is the culprit.

Unfortunately, implementing the Lustre file system using a mix of separate components from different vendors takes its toll on HPC projects in the form of wasted time, inefficiency, excessive expense, and loss of productivity. Compounding problems caused by lack of interoperability and compatibility between separate components, without end-to-end management tools and support, system administrators are on their own to troubleshoot errors. Further, system administrators constantly face tuning issues and intermittent race-conditions arising from the interaction of non-integrated servers, storage and software. This represents unacceptable and yet entirely avoidable problems that directly impact an organization’s bottom line.

Today there is an extraordinary explosion of commercial Big Data being generated by industries such as the life sciences, oil and gas, manufacturing, defense, and technical computing in general. This is driving the demand for comprehensive solution assurance, measured in industrial grade efficient, reliable, cost effective HPC storage that can handle massive amounts of sustained workload consisting of both structured and unstructured data.

Next Generation Storage

Fortunately HPC storage is evolving to meet these challenges. The industry is developing next generation scale-out storage systems that are pre-integrated, pre-tested, and designed for cost effective high productivity and reliability.

These systems are characterized by:

  • Cost-effective, efficient scalability using modular storage building blocks, including all hardware, software, networking, and the parallel file system
  • Fast concurrent data access for large numbers of users working collaboratively
  • High system robustness and resiliency to ensure high availability
  • Faster, easier system implementation made possible by factory-installed and tested rack solutions, featuring vastly reduced cable complexity – solutions are deployed in hours, not days or weeks.
  • Streamlined system expansion with the addition of turnkey embedded storage and processing nodes
  • Smaller footprint, high density configuration to reduce floor space, rack space envelopes, and power and cooling requirements
  • Efficient integrated end-to-end management through a single user interface along with integrated end-to-end support

An Innovative and Efficient Solution

The ClusterStor™ family of HPC scale-out storage solutions, introduced in 2011, provides these capabilities. ClusterStor is optimized from the disk to the file system creating the only truly engineered HPC storage solution from a single vendor.   Systems range from the ClusterStor 1500 for use by departmental level clusters to the ClusterStor 9000, specifically engineered for the largest and fastest HPC and Big Data installations on the planet.

No other HPC storage system matches ClusterStor’s proven cost effective capability to reliably sustain the fastest currently available Lustre file system I/O throughput with the least amount of hardware, while supporting installations from tens of terabytes up to tens of petabytes of data.

The key to ClusterStor capabilities is the solution’s integrated scale-out storage hardware and software architecture. The ClusterStor engineered solution packages high performance, distributed computing storage elements that encompass everything needed – from the disk enclosure and storage processing platform to the embedded operating system, file system, data protection layer, networking and advanced user interface.

At the heart of the ClusterStor scale-out storage architecture is the Scalable Storage Unit (SSU). The SSU comprehensively integrates what used to be entirely separate storage, network, and file system components into a single easy to use, rack mountable, modular building block. The SSU functions as a linear scaling unit that delivers a predictable level of performance and data storage capacity without any of the waste, expense and delay associated with legacy methods.

Overall, the resilient and consolidated ClusterStor operating environment from a single solution provider takes full advantage of the performance and scaling capabilities of the Lustre file system while eliminating problems associated with deploying separate components from different vendors. The ClusterStor approach to HPC storage boosts productivity while lowering TCO. Operational up-time and application availability are maximized through ClusterStor’s built-in high availability design and comprehensive set of distributed management services.

Pre-integrated, pre-tested, and designed for cost effective high productivity and reliability, the ClusterStor family provides resiliency engineered enterprise class storage solutions, for industrial grade HPC and Big Data applications.

Meeting the Challenge

The ascendancy of big data has catapulted HPC storage into the limelight. Xyratex a Seagate company, with its ClusterStor offerings, has risen to the challenge by creating an innovative family of storage appliances designed specifically for today’s compute-intensive operating environment that are easy to use, easy to manage and deliver the fastest results at any scale.