Supercomputing's pied piper
2008 GCN Technology Leadership Award winner Holland helps lead high-performance systems back to the front of the pack
- By Joab Jackson
- May 15, 2008
Charles Holland is deputy director of the Information Processing Techniques Office at the Defense Advanced Research Projects Agency and is a 2008 GCN Technology Leadership Award winner.
Few people have had more influence on the direction of government high-performance computing ' and computing overall ' than Charles Holland. Before others in the military or even industry, he realized how dangerously far behind the United States was falling in supercomputing prowess. Holland stressed how important keeping pace would be to national security.
MORE ON THIS TOPIC: 2008 Technology Leadership Awards home page
He also understood that high-performance computing meant more than just faster processors ' a view industry is just beginning to adopt today.
Holland is deputy director of the Information Processing Techniques Office (IPTO) at the Defense Advanced Research Projects Agency, and he oversees development in cognitive systems, language processing, sensors and other areas of computational interest. He also is program manager of High Productivity Computing Systems (HPCS), a project poised to bring supercomputing into the petascale era.
Holland has been working with the Defense Department for some time. From 2000 to 2005, he oversaw a $10 billion annual budget as deputy undersecretary of Defense for science and technology. Holland, who has a doctorate in mathematics, tackled problems that flummoxed managers with less technical savvy.
Before setting up shop at the Pentagon, he oversaw research efforts for the Navy and Air Force. In the early 1990s, he was instrumental in starting DOD's High Performance Computer Modernization program, one of the department's longest-running information technology acquisition programs.
Despite these high-profile jobs, he is perhaps best-known for HPCS. Holland helped kick off this program when he was at DOD. In 2000, he took the lead in writing a report that described how the United States was losing its edge in supercomputing and, as a result, might have been losing tactical advantages. He proved to be prescient ' a few years later, Japan commissioned the world's fastest supercomputer, one that easily outperformed the best in U.S. government or industry.
After the report was presented to Congress, Holland was asked to commission a program to get the country back to the cutting edge. The job would be immense. Every agency involved needed a supercomputer, but all of them had different uses. Holland's job was to try to get everyone together on the same technological page.
When he joined DARPA five years later, he became manager of the program he had chartered.
'The program is the largest supercomputing research-and-development program ever,' said Peter Ungaro, president and chief executive officer at Cray. 'It is definitely the driving force right now in shaping the future of HPC.'
IPTO applies the cutting edge of scientific research to national security ' in Holland's words, 'looking at those ideas that are on the far side and bringing them into the near side.'
'We concentrate our activities in areas where we think we can have a significant impact,' he said. 'We don't try everything; we pick out certain things and do that.'
With the HPCS program, for instance, Holland looked not only at improving performance ' as industry usually does ' but also productivity. During the past few years, supercomputer manufacturers have focused on how fast they can get these computers to run as measured in trillions of floating-point operations per second, or teraflops. They are starting to understand that greater horsepower is not valuable if the systems can't be controlled efficiently.
'If it takes 10 years to develop code that runs in one hour instead of nine, then you still have that nine years upfront,' he said. HPCS will look at ways not only of making machines faster but making it easier to develop code and transfer it from one system to the next.
'Productivity has been something that has been difficult to come by in very high-end systems,' Ungaro said. 'So I think [Holland] put emphasis on the areas that really need the most help. He's been out in the field, so he understands what HPC can do.'