->Skip Navigation<- National Aeronautics And Space Administration - link to http://www.nasa.gov   +Home
+Sitemap
+Feedback
+FAQs
+Other Centers
+en Espanol

 
   Home >  Life on Earth >  Using Information and Computers >  NAS Altix System


 
Life on Earth linkLife on Earth
 Introduction
link icon The Earth
link icon Improving Flight
link icon Living Things
link icon Using Information and
     Computers
 Overview
 Brahms
 CSD
 ER-2
 Kalpana Supercomputer
 Nanotechnology
arrowlink icon NAS Altix System
 PDARS
 PSRS
Humans in Space linkHumans in Space
Exploring the Universe linkExploring the Universe
About NASA AMES Research Center linkAbout NASA Ames
       Research Center



Life on Earth
NAS Altix System Off To Promising Start


Successfully meeting NASA mission goals requires a host of resources, including engineering experts, support staff, software, and supercomputers. The NASA Advanced Supercomputing (NAS) Division contributes to this pool of mission-critical resources by developing and supporting some of the largest global shared memory, single-system image (SSI) computers in the world. The most recent addition is a 512-processor SGI Altix 3000.

The Altix system is slated to be a significant engineering tool for the agency's mandate to improve the safety of future missions. The machine's ability to handle large unscheduled workloads will greatly reduce the turnaround time for some of NASA's most important applications in aeronautics, aerospace, and the Earth sciences. The Altix is also expected to improve the agency's computational modeling capability for highly parallel codes that require fast networks -- like climate and ocean modeling codes. Because it is an SSI machine (meaning all processors are seen by the computer's memory at the same time), the Altix can run a single job using almost 100 percent of the system, making modeling the entire Earth more feasible than ever.

Just the Right Ingredients
To date, over 40 applications have been transferred to the new Altix system, including many large production codes for computational fluid dynamics (CFD) such as OVERFLOW, TLNS3D, and Cart3D. The new system also serves as a testbed for developing future large, shared-memory computing systems and has already demonstrated great performance potential. "From the standpoint of reliability, stability, and features, all of the codes I'm working with can run on the Altix in production mode," says Jim Taft, a senior scientist at NAS.

Powered by a Linux operating system and Intel's 64-bit Itanium 2 processors -- one of the "hottest" performing chips currently available on the market -- the Altix system scales with relative ease. With redesigned NUMA interconnects that enable global shared memory and improved router performance, the Altix has good network characteristics--a fast connection and the hefty bandwidth required to solve NASA's large-scale technical problems like ocean and climate modeling.

Fitting the Pieces Together
Last fall, after operating just a few weeks under rigorous testing, the Altix system (initially configured with 256 processors) proved stable and robust enough to run code around the clock, with job scheduling managed by the Portable Batch Scheduler (PBS), originally developed at NASA Ames. Members of the SGI-NAS team got the new system up and running with very few problems, and computer scientists were running production codes on it almost immediately --performing four times faster per processor than NAS' largest machine, a 1,024-processor SGI Origin 3800. "The first day as a 256-processor machine, the Altix was already running at virtually the same performance as the 1,024," reports Taft. "The Altix has substantial performance potential." Without any optimization, the input/output performance of the Altix is also several times that of the 1,024-processor Origin system.

"As we got more data, we kept feeding it into the Cart3D simulations until we got better and better results," Aftosmis continued. The team had to repeatedly calculate every motion of the foam until they arrived at something that closely matched the videos. "You're always rebalancing and adjusting the calculations, and you can't do it by hand," Aftosmis says.

Improves Climate, Ocean Modeling
Because the Altix is especially well suited to handle climate and ocean modeling codes, the NAS Division is collaborating with several organizations to optimize and run their applications. One such code is the Parallel Ocean Program (POP), an ocean modeling code from Los Alamos National Laboratory. Another climate, ocean, land, and ice model from the National Center for Atmospheric Research (NCAR), called the Community Climate System Model (CCSM) is also being run on the new system. NASA's Jet Propulsion Laboratory is also taking advantage of the Altix to run the ECCO code (Estimating the Circulation and Climate of the Ocean), a coupled ocean and climate model. According to Taft, all these climate and ocean modeling codes are getting excellent performance results on the Altix system.

Leading and maintaining the development effort of large shared-memory systems is one of the division's ongoing goals -- upgrading to a 1,024-processor system, one of the division's near-term plans for the Altix, will help achieve this goal. In addition to increasing the number of processors, an imminent goal for the Altix is to optimize the system's performance. Says NAS research scientist Bob Ciotti: "Having established the viability of the Altix, we will further improve its reliability, scalability, and aggregate performance for NASA applications. Specifically, we will develop interfaces that enable an application to share memory, while breaking the requirement that the system be run as a `single-system image'. Demonstrating that this approach is feasible will provide valuable information in the development of future computing systems."

-- Holly Amundson