GAO: DOE supercomputers underused, underscrutinized

The Energy Department is not getting enough use out of its 42 supercomputers and needs to take a closer look at its plans to buy five more machines during the next two years, the General Accounting Office said in a new report.

The study, requested by House Budget Committee Chairman Rep. John Kasich (R-Ohio), said the DOE systems, mostly scalable parallel supercomputers purchased since 1994, run at an average of 59 percent of their processing capacity. Meanwhile, GAO charged, decisions about future investments in the technology— including the $4 billion Accelerated Strategic Computing Initiative (ASCI)— are not being given the same scrutiny as DOE's administrative systems that is required by the Clinger-Cohen Act of 1996.

In a written response included with the July 17 report, DOE undersecretary Ernest Moniz said GAO's measure of capacity in the systems— the amount of time processors were idle— did not account for technological limitations of the world's fastest computers. In addition, he said, investments in scientific computers require "different management approaches'' than other systems.

John Pike, a defense analyst with the Federation of American Scientists, said the GAO report reflects a general view the agency has taken that DOE needs to manage its programs more centrally.

"It's an awful lot of money, and when you look at it [it] looks like you have lot of big computers sitting around and doing not very much,'' he said. "These are not experimental supercomputers; [DOE is] buying them to do things. [DOE users] are basically trying to grab the biggest computer they can get.''

In the past four years, DOE has spent $301.5 million on new supercomputers for its laboratories, plus $526 million to run them. The ASCI program, which is the primary supercomputer user, relies on the systems for developing the capability to simulate nuclear weapons tests, for research into global climate change and decoding the human genome.

But GAO said the department does not monitor how much the machines are used and thus cannot take unused processing capacity into account when deciding whether to buy new ones. "If DOE could improve its utilization rate by 10 to 15 percent overall, it could save tens of millions of dollars in new acquisition costs,'' the report said.

Three years ago DOE tried to eliminate extra paperwork by canceling a policy that required labs to submit detailed plans for their computer acquisitions. As a result, GAO said, "even the most expensive systems are not necessarily visible to the department's information technology managers,'' contrary to Clinger-Cohen.

Moniz countered that the purchases do get enough scrutiny from the department and the Office of Management and Budget. He added, however, scientific computing equipment— "which pushes the state of the art'' and has a limited user base— cannot be judged by the same criteria as administrative systems built with "proven technology.''

Meanwhile, Moniz said in the report, the amount of time each supercomputer spends processing data is limited by memory size, internal bandwidth and other technical constraints. For example, he wrote, moving data into and out of memory requires "significant computer time, during which the processors are mostly idle'' but not available for running other programs. Furthermore, he said, new, experimental systems are not designed to be widely used because they are "less stable'' than "production'' machines.