Skip all navigation and jump to content Jump to site navigation Jump to section navigation.
NASA Logo - Goddard Space Flight Center + Visit NASA.gov
Computational and Information Sciences & Technology Office banner

  +Home

 

CISTO News
HOME
ARCHIVES

Download a PDF version of the newsletter:

+ CISTO News Spring 2006

High-Performance Computing

 

NCCS to install Next-Generation Supercomputer

The NASA Center for Computational Sciences (NCCS) will soon install the first stage of its next-generation supercomputer, a Linux Networx Custom Supersystem. A 128-node “base unit” is expected to arrive at Goddard Space Flight Center (GSFC) in late June. While providing an initial peak performance of 3.3 teraflops, the system could scale to nearly 40 teraflops in its full configuration.

Image at right: A Linux Networx Custom Supersystem will be the next-generation supercomputer for the NASA Center for Computational Sciences. The first stage of the cluster, providing 3.3 teraflops of peak computing power in five closet-sized cabinets, will arrive in June (Photo credit: Linux Networx).

NCCS system integrator Computer Sciences Corp. (CSC) selected the cluster through a rigorous procurement process involving a nationally distributed Request for Proposals. Designed and optimized for each customer, a Custom Supersystem incorporates a variety of technologies from vendor partners.

Each node in the NCCS base unit contains two Intel dual-core, 3.2 GHz Xeon-Dempsey chips and 4 gigabytes of memory. The system thus has a total of 512 processing units and 512 gigabytes of memory. Its nodes are linked by a 10-gigabit-per-second Infiniband Network from Silverstorm Technologies. Tightly integrated with the compute nodes is a high-performance storage subsystem with 60 terabytes of raw Data Direct Networks storage running the IBM General Parallel File System.

Linux Networx Custom Supersystem

 

“I am very excited to welcome Linux Networx and their partners to the NCCS team,” said Phil Webster, CISTO’s Lead for High-Performance Computing. “Their high-performance computing expertise and ability to integrate the latest technologies will provide our Earth and space science users with an enhanced processing and data analysis capability for their research.”

The base unit’s peak performance is 100 gigaflops higher than the NCCS’s retiring 1,392-processor HP AlphaServer SC45, with a footprint one-tenth that of the older machine. By early July, the NCCS expects to make the cluster available to pioneer users and then begin a 30-day acceptance period that will include baseline benchmarking. While undergoing evaluation, the base unit will run side-by-side with the SC45. NCCS plans call for adding one or two 256-node “scalable units” to the Custom Supersystem by the end of calendar year 2006, increasing aggregate performance to 9.9 or 16.5 teraflops peak. As the system expands, portions of the SC45 will be shut down to accommodate the power and cooling needs.

The Custom Supersystem will provide capacity computing for the NCCS, with most jobs using up to 64 processors. However, it also will be available for capability runs, i.e., for applications that require and can scale to larger numbers of processors. The 1,152-processor SGI Altix 3700 BX2 system, which has twice as much memory per processor, will continue to be the NCCS’s main capability computing platform.

“One of the primary architectural goals of the new cluster environment was to make the transition for the user community as smooth as possible,” said Dan Duffy, Lead Architect for CSC. “We purposefully chose a design that presents a consistent user environment between the Custom Supersystem and the Altix. Users will have access to the same set of tools, modules, and home file systems on both platforms.” For instance, the cluster will have compilers from Intel and PGI, run the PBS Pro batch scheduler, and have both MPI and OpenMP, along with other valuable software tools such as TotalView, MATLAB, and IDL. Moreover, programmers from the NCCS and GSFC’s Software Integration and Visualization Office (SIVO) will assist users in porting and optimizing their codes.

The Custom Supersystem approach also includes the ability to add special processing nodes for analysis and visualization as well as application-accelerator nodes using field-programmable gate arrays, more generally known as FPGAs. The NCCS is evaluating such technologies for its system and will consider user input on what options would be of broadest benefit.

http://nccs.nasa.gov
http://www.linuxnetworx.com



USAGov logo + Privacy Policy and Important Notices
+ Sciences and Exploration Directorate
+ CISTO
NASA Curator: Pamela Ricks
NASA Official: Phil Webster, CISTO Chief
Last Updated: Thursday, 06-Dec-2007 10:41:56 EST