Alliance-enabled Capability Computing

You've heard the phrase and you're interested. Exactly what do NCSA staff mean when they use the words "capability computing"? Who is eligible to request access to such resources and how can this request be made?

The Alliance has three goals in the area of capability computing: provide the highest level of computing technology to academic researchers, provide operational support for the four classes of capability computing projects, and develop new software tools, libraries, and software to make capability computing easier and more efficient.

Capability Computing Support through the Alliance

NCSA plans to enable capability computing by putting the highest level of computing technology at the disposal of academic researchers by offering operational support for:

• breakthrough "hero" simulations—projects so memory, CPU, data, or I/O intensive that they require dedicated large resources for multiple days or weeks
• high throughput simulations—uncoupled simulations where rapid turnaround is required
• rapid turnaround supercomputing—simulations that are too large for a user's desktop system but that are not groundbreaking in their technical requirements or sheer numbers
• highly interactive visual applications—analysis and visualizations that are too large to be moved to a user's desktop system

The availability of new NCSA dedicated processing power make cutting-edge science possible without sacrificing time reserved for smaller computing jobs that are important to the general research community.

Did a Project Provide "Proof of Concept"?

NCSA was pleased to provide the hero simulation resources needed for the recent successful demonstration of real-time numerical weather forecasts by the Center for Analysis and Prediction of Storms (CAPS) at the University of Oklahoma. Kelvin Droegemeier and other CAPS members were given exclusive access to a full 128-processor Origin dedicated periods during a two-week period in January 1999.

The real-time nature of the forecasts presented a unique challenge to NCSA. Normally, Origin2000 users are scheduled into a queue, and their jobs run whenever there are enough cycles available, which could be any time during the day or night. However, with real-time calculations, the Origin's computing power was needed at specific times. CAPS researchers generated 10 daily forecasts for 7 days straight, which required: 2 gigabytes of observational data per forecast; 2 million computational cells; 1 million billion floating point operations; and 6 billion floating point operations per second The computational demands of digesting 1.5 gigabytes of Doppler data and then calculating these variables explains why Droegemeier's team needed all 128 nodes of NCSA's Origin2000 to run their model in real time. This spring Droegemeier is using an entire 256-processor Origin for dedicated periods over seven weeks in an effort to create high-resolution forecasts in localized areas across the eastern two-thirds of the U.S.

How Can My Project Be Considered?

Determinations about allocating resources for capability computing needs are made by the NCSA Computer Policy Committee, co-chaired by NCSA Senior Associate Director John Towns (Scientific Computing and Visualization) and Associate Director Jeff Terstriep (Computing and Communications).

In order to make necessary arrangements, written requests must be made at least 6 weeks prior to the suggested period of the capability computing run. Droegemeier's request [PDF format] is a good model to follow. A request should contain an overview of the project, a detailed list of requested NCSA resources (e.g., project/allocation information, CPU and duration needed, number of processors, mass storage needs, networking), and contact information. This information can be provided in an email message or as a document. Send the information to Towns.

For More Information On …

• capability computing, read the article by Larry Smarr in R&D Magazine this summer.
• capability computing, review the presentation materials and video archive from the conference sponsored by SGI and NCSA in May 1999
• Kelvin Droegemeier's work, review the Access article that reported on his projects.
• NCSA high-performance systems, check the NCSA High-performance Systems information.

-- Ginny Hudak-David
06/99