Architecting HPC Data Storage Solutions

By Ken Claffey

December 19, 2011

Innovation has been the cornerstone of success in our heritage in the data storage industry for the last twenty-five years.   About two years ago, Xyratex initiated an investigation into additional market opportunities for enterprise class data storage solutions.  Our research yielded interesting data points that aligned with the strengths of Xyratex.  We discovered that not only was the High Performance Computing (HPC) a high growth area for storage, it also represented a dynamic market opportunity with a substantial need for better data storage design. We also learned that the way data storage was being implemented at many of these sites was unduly complicated in terms of initial installation, performance optimization and ongoing management.  

The performance requirements of HPC environments with ever-larger compute clusters have placed unprecedented demands on the I/O performance of supporting storage systems. Storage I/O has become the performance bottleneck for many HPC clusters, as traditional storage systems cannot keep pace with the increasing scale of these deployments. Clearly, a state-of-the-art, scalable HPC storage solution is required. What would a high-performance storage architecture that offers a new level of component integration and performance optimization in HPC environments look like?

The HPC market is going through a paradigm shift. The adoption of low-cost, Linux-based clusters that offer significant computing performance and the ability to run a wide array of applications has extended HPC’s reach from its roots in scientific laboratories to smaller workgroups and departments across a broad range of industrial segments, from biotechnology and cloud computing, to manufacturing sectors such as aeronautics, automotive, and energy.

With dramatic drops in server prices, the introduction of multi-core processors, and the availability of high-performance network interconnects, proprietary monolithic systems have given way to commodity scale-out deployments. Users wanting to leverage the proven benefits of HPC can configure hundreds, even thousands, of low-cost servers into clusters that deliver aggregate compute power traditionally only available in supercomputing environments.

As HPC architecture has evolved, there has been a fundamental change in the type of data managed in clustered systems. Many new deployments require large amounts of unstructured data to be processed. Managing the proliferation of digital data – documents, images, video, and other formats, places a premium on high-throughput, high-availability storage. The explosive growth of large data has created a demand for storage systems that deliver superior I/O performance. However, technical limitations in traditional storage technology have prevented these systems from being optimized for I/O throughput. Performance bottlenecks occur when legacy storage systems cannot balance I/O loads or keep up with high-performance compute clusters that scale linearly as new nodes are added.

Despite the obvious advantage in application performance offered by HPC cluster environments, the difficulty in optimizing traditional storage systems for I/O throughput, combined with architectural complexities, integration challenges, and system cost have been barriers to wider adoption of clustered storage solutions in industrial settings. With the introduction of the Xyratex ClusterStor 3000™, a fully integrated, Lustre® based storage cluster solution, these challenges have been met.

Scale-Out Storage Architecture Overview

At the heart of the ClusterStor 3000 is the Scale-Out Storage Architecture, an advanced storage design that provides higher density and system reliability at lower cost. Fully integrating the Scale-Out Storage Architecture with the Lustre file system creates a next-generation HPC storage solution that delivers simplified system installation and operation, optimized HPC performance, and non-disruptive cluster expansion.

Limitations of Traditional Server/Storage Systems

Traditional storage clusters are made up of a number of disparate building blocks, including servers to run the file system and associated software, a high-speed storage interconnect, such as InfiniBand, connected to a RAID controller, and a high-density storage system housing the disk. Each subsystem in this hardware/software combination adds complexity and potential bottlenecks in terms of balancing I/O, reliability, and system scalability.

Scale-Out Storage Architecture – Distributed I/O Processing at Scale

The ClusterStor Scale-Out Storage Architecture is a consolidated hardware and software environment, designed from the ground up to uniquely address the demand for  Lustre scalability and performance, which features an integrated Scalable Storage Unit (SSU). Each SSU supports two industry-standard x86 Embedded Server Modules (ESMs), which connect directly through a common midplane to all drives in the SSU, and share a redundant high-speed interconnect across the midplane for failover services. The ESMs are capable of running industry-standard Linux distributions, and each module has its own dedicated CPU, memory, network, and storage connectivity. When new SSUs are added to the cluster, performance scales linearly as incremental processing network connectivity and storage media are added with each unit. This modular design removes the performance limitation of traditional models in which servers or RAID heads quickly become the bottleneck as more drives are added to the cluster. The Scale-Out Storage Architecture combines enclosure and server enhancements with software stack optimizations to deliver balanced I/O performance (even on large data workloads), and outperform traditional storage topologies by adding easy-to-install, modular SSUs that scale ESMs as HPC storage scales, distributing I/O processing throughout the solution.

Overcoming Lustre’s Challenges

Despite the obvious advantages of using Lustre as a clustered file system in HPC deployments, legacy installations have shown limitations and challenges to users. Integrating Lustre into an HPC solution that overcomes these challenges drove the development and design of the ClusterStor 3000.

Installation and Administration

Lustre is traditionally installed and administered from a command-line interface, requiring fairly advanced knowledge of Linux administration and experience provisioning clusters and managing distributed nodes. System administrators may lack this knowledge or not understand the complexities of the Lustre software, requiring them to rely heavily on professional services and external support to deploy and administer the file system. Additionally, integrating Lustre into a storage system with heterogeneous hardware adds system and support complexity. It can take weeks to get the system up and running reliably, with additional effort required to tune system performance or upgrade large systems without interruption of service

ClusterStor Approach:

ClusterStor systems come installed and configured from the factory, fully optimized to work with Lustre, and ready to deploy right from the loading dock. This innovative HPC storage design leaves behind the complexities of other systems that integrate third-party equipment and require tuning of file system performance in a heterogeneous, multi-vendor infrastructure.

Additionally, the ClusterStor 3000 includes a browser-based GUI that features intuitive wizards to guide administrators though automated Lustre network configuration and user authentication, greatly easing cluster setup. Once the file system is running, the ClusterStor Manager is used to administer the Lustre cluster (including failover and other tasks), monitor operations, analyze performance, gather statistics, generate reports and update the system non-disruptively.

Data Redundancy and Protection

Lustre maintains the integrity of data in the file system, but it does not back up data. Lustre relies exclusively on the redundancy of backing storage devices. To protect data from loss, Lustre users must attach external RAID storage.

ClusterStor Approach:

ClusterStor’s Scale-Out Storage Architecture uses a highly-available RAID 6 array to protect system data and provide optimal fault tolerance and performance. The Lustre file system’s Object Storage Servers (OSSs) are configured for active-active failover, enabling all logical drives to be seen by both OSSs at all times. If one OSS node fails, the active OSS takes over OST management and operations of the failed node. In normal non-failure mode, I/O load is balanced between the OSSs.

System Monitoring and Troubleshooting

Lustre generates events to indicate changing states and conditions in the file system, but it does not provide comprehensive file system monitoring. In the event of file system errors, log descriptions may clearly indicate the problem, but many entries are difficult to interpret without Lustre troubleshooting experience. Additionally, Lustre does not isolate file system errors from faults in the client/server hardware, network interconnect, or backing storage system. In a large-scale file system, determining root cause may be difficult.

ClusterStor Approach:

ClusterStor Manager provides a unified system view of all Lustre nodes, with combined syslogs available to monitor events in the cluster. To enhance the usability of the Lustre logs, filtering and record sorting is available. Additionally, the ClusterStor Manager rolls up log information into diagnostic payloads, ensuring that support engineers have all the information needed to help customers diagnose and resolve issues quickly.

As part of the ClusterStor 3000, Lustre is deployed on a hardware platform offering sophisticated fault isolation to pinpoint the location of a failure, with easy field serviceability to quickly replace failed parts and system wide high-availability, enabling uninterrupted cluster operations if a component fails.

ClusterStor 3000 delivers:

Industry-leading IOPS performance and solution bandwidth – distributed I/O processing and balanced workloads as the system scales linearly

Faster, easier system implementation and reliable expansion – factory-installed and tested components, reduced cable complexity with solution setup in hours, not days or weeks, plus streamlined system expansion with turnkey Scalable Storage Units (additional storage and processing nodes)

Fully integrated hardware and software platform – smaller solution footprint (fewer components, lower cost, and improved investment protection)

Simplified cluster management – ClusterStor Manager automates configuration and node provisioning, and provides performance statistics, system snapshots, and syslog reporting

Extreme reliability, availability, and serviceability – No single point of failure and redundancy built in throughout.

Visit www.xyratex.com  to learn more about the ClusterStor HPC data storage solution.

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion XL — were added to the benchmark suite as MLPerf continues Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

MLPerf Inference 4.0 Results Showcase GenAI; Nvidia Still Dominates

March 28, 2024

There were no startling surprises in the latest MLPerf Inference benchmark (4.0) results released yesterday. Two new workloads — Llama 2 and Stable Diffusion Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire