The Coming of a Golden Age:

Lessons Learned from
Capability Computing Conference

For the first time, the complexity of computational research problems being tackled is starting to approach the complexity of nature. Greater computational power and the aggregation of computer resources makes solving complex problems more possible.

A score of research projects at NCSA use unprecedented computational resources - more than 50% of the center's capacity annually. Such projects, examples of so-called capability computing, are simulations so memory, CPU, data, or I/O intensive that they require the dedication of a large resource for days to weeks. NCSA Director Larry Smarr considers the capability computing metric to be the routine execution of a problem 100 times larger or longer running than is possible on a modern desktop system. Kelvin Droegemeier's weather modeling and prediction work is an example. His team's model was successfully demonstrated at the American Meteorological Society meeting in January 1999 but required a dedicated 128-processor Origin during a one-week period.

Such complex and multifaceted problems can be tackled because of the availability of cache coherent, non-uniform memory architecture (ccNUMA) systems, such as SGI's Origin2000. Distributed shared memory (DSM) found on the Origin2000 is the next generation of computer architecture beyond massively parallel computers. DSM machines combine large parallelism with huge shared memories.

In May 1999, SGI and NCSA jointly sponsored a conference and workshop on capability computing targeting the Origin2000 system. The conference goal was to offer suggestions on how to optimize a wide variety of disciplinary applications to take full advantage of the Origin2000's ccNUMA configuration. Presentations on application tuning, optimization, tools, and performance were featured during the two-day workshop that followed the conference.

Academic researchers' sessions featured a series of diverse and creative solutions to programming challenges to take advantage of ccNUMA systems, heralding the arrival of a "golden age of applications" according to Smarr. In the conference wrap-up, Smarr observed two significant trends from the presentations by industry professionals and computational researchers. The first is that computation models are and will continue to be hybrid, meaning that programmers will take the best aspects of languages or approaches and marry them to accomplish project goals. Second, in keeping with the spirit of the hybrid model, Smarr noted that there is no single "how to" to make ccNUMA systems work optimally. Researchers will have to be innovative and creative to accomplish their computing goals.

NCSA and the Alliance are ready to help make such goals reality. Researchers should contact the NCSA Consulting Office for assistance with their allocation. §

--Ginny Hudak-David