Not logged in » Login
Jan 14 2016

PRIMEFLEX for HPC: The Right Choice for High-Performance Technical Computing

/data/www/ctec-live/application/public/media/images/blogimages/graphics/PF_for_HPC.png

Once restricted to areas like scientific research, oil and gas exploration or car development, techniques like simulation or CAD/CAE have long since left their original turf and become commonplace. Today, you can hardly find a product or service that has not been developed – or at least significantly improved – without the help of computer programs: bicycles are tested in virtual wind tunnels, and smartphones are exposed to the possible effects of gravity when hitting simulated surfaces instead of real ones. Even the weather forecasters on your favorite news channel now present CGI animations that show in detail how the sun, wind and rain will behave in the next 24 hours. As a result, more and more companies need platforms that enable them to perform high-performance computing (HPC) tasks. FUJITSU CELSIUS workstations as well as the FUJITSU Integrated System PRIMEFLEX for HPC fulfill their demands.

When general audiences first learned about high-end computer graphics and CAD in the mid-to-late 90s, the required software packages – such as PowerAnimator – sold for five-digit amounts and typically ran on MIPS-based hardware stacks that cost just as much. Two decades later, five of the world's ten fastest supercomputers employ x64 architectures with matching co-processors, and the percentage is even higher in the 'lower' ranks. In short, most high-performance computing (HPC) and supercomputing tasks that formerly required special exotic CPU models may nowadays be carried out using off-the-shelf processors designed for standard servers – a fact that makes building and implementing HPC solutions a lot easier than it used to be. This shift to a more affordable platform has in turn spawned a process one might call democratization: these days, more and more companies – from startups building racing bikes or creating graphics-heavy computer games through architectural firms and industrial designers to global teams developing reliable models for financial analysis or climate research – seek to benefit from the massive power and performance boost that dedicated HPC solutions will bring to their line of business. The growth in popularity will help ICT vendors create new revenue streams – an opportunity neither Fujitsu nor our channel partners can afford to let go. However, when it comes to deploying such solutions there's typically one major obstacle: existing hard- and software configurations for HPC are often complicated to set up, difficult to manage and arduous to use. A modern, customer-friendly HPC implementation has to solve these issues.

This is exactly the kind of challenge we try to address with our powerful, easy-to-use-and-manage PRIMEFLEX family of integrated systems, which cover just about any complex task IT departments may face today – from server and desktop virtualization to analyzing Big Data volumes and managing cloud environments. The HPC variant – full name: FUJITSU Integrated System PRIMEFLEX for HPC – combines CELSIUS workstations, PRIMERGY servers and optionally ETERNUS storage systems into unified platforms that can be obtained from a single vendor and have already been successfully deployed across various industries. More specifically, the workstations connect to HPC Clusters made up of Fujitsu blade, rack or scale-out servers and disk storage arrays, and all components are linked together by a specially developed software stack – namely, a cluster suite, a file system, and an application desktop – so they may be easily modified to suit industry- and/or company-specific requirements.

Architecture
PRIMEFLEX for HPC follows the same design principles as any other member of the product family. That means, it is delivered as a pre-configured, pre-integrated and pre-tested combination of the required hardware components (servers, storage, workstations, and networking gear) and software packages (for management, virtualization, automation/orchestration, plus database and apps). PRIMEFLEX for HPC is available in two 'dosage forms' – either as ready-to-run application appliances or as reference configurations that may be tuned at the customer's site. Otherwise, both variants are similar in structure and may be optimized to perform certain tasks or, better still, run special application software, as you will see in the examples below.

Building Blocks – Hardware
Fujitsu's scalable HPC architecture is flexible enough to respond to the most diverse requirements; hardware platforms for implementation range from high-end CELSIUS workstations to PRIMERGY industry-standard servers, all of which are built around Intel® Core™ or Xeon® processors.1 More specifically, PRIMEFLEX for HPC solutions may include several of the following components:

  • CELSIUS Workstations: CELSIUS C740 rack workstations or whisper-quiet dual-processor R940 desktop workstations

PRIMERGY CX2570 M1 and RX350 S8 servers as well as the CELSIUS R940 may additionally be equipped with Intel® Xeon Phi™ co-processors.

Fig. 1

Fig. 1: The combined building blocks of PRIMEFLEX for HPC

Building Blocks – Software
The operating system layer for these platforms typically consists of a popular Linux distribution, e.g. from Red Hat, SUSE, or the CentOS project. On top of that, we add these software components:

  • The FUJITSU Software HPC Cluster Suite V3.1: The HPC Cluster Suite (HCS) is a purpose-built software stack designed to simplify deploying, managing and using HPC clusters. HCS includes the HPC Gateway, a Web-based end-user interface customized for HPC usage (see below).
  • The Fujitsu Exabyte File System (FEFS): An offspring of the open-source, parallel distributed Lustre file system, FEFS was designed to fulfill the highest performance, scalability, I/O and availability requirements, allowing for up to 8 exabyte storage capacity and 1 TB/s data transfer rates along with fair-share and best effort QoS as well as transparent failover to redundant hardware and uninterrupted file system services. FEFS also runs on Fujitsu's K computer, which first appeared on the Top500 list of supercomputers in 2010 and has ranked among the ten fastest systems ever since, and its cousin PRIMEHPC FX100, which was designed to enable "petascale computing" and first introduced in late 2014. (For more details, please refer to the data sheet.)
  • The HPC Gateway Application Desktop: As noted above, the HPC Gateway is a core component of HCS that simplifies all aspects of HPC work management with integrated functions for file management, application execution and result monitoring. Its web interface uses a familiar desktop layout to ensure the user experience is both comfortable and intuitive. As a result, preparing, launching and monitoring HPC jobs becomes both very easy and highly effective, an effect that will appeal to new or occasional HPC users and skilled ones alike. Main functions of the HPC Gateway include application and file explorers, a job scheduler, and a workflow editor. Combined with the pre-built packages from the Application Catalogue, these provide the most productive and coherent HPC workplace in the market today.
  • The HPC Gateway Application Catalogue was developed to improve workflows and increase user productivity. It is populated with a set of intelligent application processes of common ISV and Open Source Software applications. These validated processes have been developed with end users and their need for robustness, visibility and operability in mind. Core features include functions such as license checking, the ability to directly show application results on the web desktop, and automatic reruns, e.g. the ability to restart applications or jobs in case of a failure without intervention from an administrator. More application processes can easily be added to the cluster using either the workflow editor or the built-in application on-boarding mechanism for scripts. The data workflow feature enables any type of HPC business application process to be easily automated. Best-practice processes make individuals more effective in their domain of expertise since they are using HPC in terms of the business purpose.

Appliances and Reference Configurations
As noted above, PRIMEFLEX for HPC solutions are either available as application appliances or as reference configurations. For user groups who have not previously worked in HPC environments, the application appliances are usually a good starting point, since they combine all required elements in a neat package, as shown at the top of this page. This includes every bit part, from workstations and compute nodes through interconnects to PSUs and racks on the hardware side and from the Web-based GUI to management software and predefined industry standard applications. For example, we offer configurations designed to support ANSYS® Fluent® and ANSYS® CFX®, two of the most popular simulation programs that solve problems in fluid dynamics and help shape prototypes and products in the aeronautic and automotive industries. More specifically, ANSYS software enables manufacturers in the field to study the interaction of their products within the complete system without the need for a physical testbed. Likewise, they allow for examining the behavior of those parts under various operating conditions. In these use cases, HPC scalability helps to create more detailed and realistic models to be computed with ANSYS applications within an invariably limited timeframe. HPC throughput capability then allows design of experiment (DoE) and robust design ensemble methods, creating several hundreds of jobs, to be run consistently on each project with integrated ANSYS process tools.

While the power of ANSYS applications is fascinating, building and further optimizing matching HPC solutions from scratch is a time-consuming, expensive and risky endeavor that may not yield the desired results. By contrast, Fujitsu's 'appliance approach' ensures that all parts fit together and work as one. Moreover, it eliminates obstacles regarding operational performance and user productivity: In traditional HPC environments, the latter often suffers because users spend more time in dealing with their equipment than carrying out simulations and analyzing the results. The Fujitsu HPC Gateway does away with such issues by providing an intuitive, easy-to-use interface slightly reminiscent of a Windows machine or Mac. Simplifying HPC in such a way will have a transformative impact not just for small project teams, but for entire companies as it becomes a standard for more engineers, designers, and technicians. Businesses can then apply this power to expand exploration of a product's design and to increase performance, quality, reliability, and ultimately innovation. As with all PRIMEFLEX solutions, the complete package also includes support and services. Similar solutions exist for a wide variety of other simulation, visualization, design and modeling programs such as ANSYS CFD, Autodesk VRED, and COMSOL Multiphysics.

For detailed information, please visit our HPC microsite or check out our set of industry solutions for the automotive, construction, educational, manufacturing and energy sectors. And don't forget to take a look at our webcast recording.

[1] Intel, the Intel logo, Xeon, and Xeon Inside are trademarks or registered trademarks of Intel Corporation in the U.S. and/or other countries.

Natalie Walther

Maximilian Brandtner

 

About the Author:

Natalie Walther

Global Workstation Product Marketing Manager, Fujitsu

About the second Author:

Maximilian Brandtner

Senior Product Marketing Manager at Fujitsu CEMEA&I

SHARE

Comments on this article

No comments yet.

Please Login to leave a comment.

X

Please login

Please log in with your Fujitsu Partner Account.

Login


» Forgot password

Register now

If you do not have a Fujitsu Partner Account, please register for a new account.

» Register now