by
Veronica Fowler
Of The Register's Ames Bureau
Ames, Iowa -- Which is faster? A tricycle or a jet?
The answer to the riddle is the key behind an Ames Laboratory computer innovation that has been hailed as one of the most significant technologies developed this year.
The Ames Laboratory at Iowa State University has developed a way to evaluate computer systems, and it is credited with revolutionizing the way computers are compared to one another.
It's particularly important as more and more researchers and businesses are spending millions of dollars on computer systems for which there is relatively little consumer information.
The Ames Laboratory creation, called SLALOM, recently was named by Research and Development Magazine as among the 100 most significant technologies developed in the last year. SLALOM, which stands for Scalable Language-independent Ames Laboratory One-minute Measurement, got its name because like the slalom races, it, too, is a timed race through obstacles.
While there have been dozens of other "benchmarks" -- computer lingo for methods of evaluating computer speed -- SLALOM is receiving praise because it tackles the evaluations in a new way. Instead of trying to figure out how long it takes to do a certain problem, SLALOM figures out how much the computer can do in a given time.
John Gustafson, an Ames Laboratory computational scientist, says it can be misleading to try to figure out how fast a computer can work simply by giving it a task and timing it. It's like trying to figure out which is faster, a trike or a jet. The jet might be faster for a distance of 100 miles, but the trike is faster for 100 feet when you consider you don't have to rev its engines; get it on the runway and into the air. On the other hand, the trike could never make 100 miles. It's a poor comparison, Gustafson says.
You'll get a more accurate picture of speed, says Gustafson, if you look at how far the trike or jet can go in, say, one hour.
Another way of looking at the difference between SLALOM and other benchmarks, says Gustafson, is that it has been "like proposing a new Olympic event called the 10-second dash."
Gustafson says SLALOM also has tremendous potential because it can compare nearly every sort of computer, from a personal computer to a complicated supercomputer.
"With SLALOM, you can compare a Macintosh personal computer to the fastest CRAY supercomputer and do justice to both," says Gustafson.
SLALOM also has the advantage of never becoming outdated, says Gustafson. Each year, computers increase their speed in solving problems at a breathtaking rate. Eventually, computers get so fast that the problems used to evaluate them can be solved too quickly, skewing evaluations. "It takes so little time that funny things start to affect it, like how long the problem took to lead into the computer."
Gustafson says having a computer evaluation system that is up-to-date and accurate is essential as researchers each year spend millions and even billions of dollars on computer systems. Too often, he says, researchers try to figure out which computer systems to buy without the basic consumer information.
Right now, even though there's a patent pending, SLALOM is available to anyone, but is geared for scientific researchers.
Computer benchmarking has proliferated in recent years. Some computer experts have been critical of many benchmarks, accusing those who developed them of giving out inflated evaluations of particular manufacturers.
Yet it is supercomputers that are essential for much research, including weapons development and even the simulation of automobile crashes.
It was that inaccuracy and lack of credibility that prompted Gustafson to develop SLALOM in the first place. He had worked several years with high-tech companies in Oregon and became frustrated with the dearth of accurate benchmarks. He came to the Ames Laboratory in 1989 with the goal of developing SLALOM. "I saw purchases being made when there wasn't good information available."
SLALOM might not only help evaluate computers, but it might also help revise the way they're designed. Gustafson says computer engineers "have been asking the wrong questions all this time." Computer designers have been striving to make computers solve problems in less time.
But they could make computers more useful if they aimed for getting more work or better results out of a given amount of time, rather than just reducing the time.
Contacts:
John Gustafson
gus@scl.ameslab.gov
Skip Derra, +1-515-294-4917, ISU News Service
The URL for this document is http://www.scl.ameslab.gov
Revised
Pages prepared by Maria E. Blanco.