Report: Commercial supercomputing software is scarce
As a result, U.S. companies often do not have the HPC application software they require.
Industry is not producing enough commercial supercomputing software to keep American companies competitive, according to a study by the nonprofit Council on Competitiveness.
The first-ever report on the state of application software for high-performance computing (HPC) systems shows that market forces and technical challenges in recent years have led software vendors to turn their focus away from creating new HPC application software. As a result, U.S. companies often do not have the HPC application software they require.
The study, released March 7, cites several recommendations geared toward the government, including sustained, multiyear funding for researchers to develop software and algorithms.
The findings grew out of a July 2005 workshop sponsored by the council, the Ohio Supercomputer Center and the Energy Department’s Office of Science. Participants in the daylong workshop on the state of HPC software included HPC users, software vendors, university researchers, hardware vendors, and government scientists and engineers.
Today, Suzy Tichenor, vice president of the council, said the government is in a position to be a large part of the solution to the problem.
“I think there is an opportunity here for the government and the private sector to work together to help to overcome a challenge that is actually in the private sector," she said. "One area that we know that the government is very, very strong in is its expertise in using high-performance computing. We think that that expertise can be leveraged to help industry address some of the problems that it has.”
Tichenor added that one of the government’s best skills is creating more mature code.
“The government could encourage the national labs to engage in partnerships with industry to accelerate the transfer of this expertise,” she said.
As part of a series of panels, several government representatives addressed the hurdles in producing new commercial software.
“Start digging as to where these software packages came from; an awful lot of them have their roots in government-funded development that was commercialized,” said Phillip Colella, senior staff scientist at Lawrence Berkeley National Laboratory. But “one of the things, in my view, that is institutionally different now than it was 30 years ago or even 20 years ago: the amount of resources going into those kinds of activities has been steadily drying up.”
In each case where software made the transition from research to production-quality HPC applications, “there was a long-term investment by the government to give these packages a chance to mature,” added Al Geist, a corporate fellow at Oak Ridge National Laboratory.
Other recommendations for the government include revising visa requirements so that software vendors can attract the talent they need and monitoring foreign acquisitions of key software vendors. Although most vendors have small sales and capitalization, their codes may be indispensable to U.S. industry and national security.
“Should a foreign entity decide to purchase any of these [software vendors] and remove their software from the market, the public and private sectors potentially could be denied codes that they use to solve even their most basic problems,” the report states.
The study outlined a road map for software vendors, universities and government research establishments to follow in addressing these barriers to commercialization of supercomputing software.
NEXT STORY: Trekkin' around FOSE