NEVER do today what you can put off until tomorrow. It's every slacker's motto, but in the world of high-performance computing it can make perfect sense. If you have a particularly tough calculation to perform, it pays to wait until a better computer comes along.
This conclusion, which will be music to the ears of procrastinators the world over, comes from a team led by Chris Gottbrath at the University of Arizona's Steward Observatory. "By starting the calculation later, when your money can stretch to buying a more efficient computer, you'll finish earlier," says Gottbrath.
Researchers in disciplines from cryptography to climatology are pushing the limits of even the best supercomputers as they tackle ever more difficult computational challenges. Computing power increases all the time, however, and Gottbrath and his colleagues realised that there might be computational tasks for which procrastination pays. "In the light of this, we wondered whether, if you had a limited budget for computers, you ought to wait before starting," says Gottbrath.
To find out, they constructed a simple mathematical model based upon Moore's law. Propounded in 1965 by Gordon Moore, one of the founders of the computer chip maker Intel, this states that the computational power available at a particular price doubles roughly every 18 months.
What Gottbrath and his colleagues found was that any calculation estimated to take longer than 26 months to run on the best hardware available today is not worth starting. "If it takes less than 26 months, our advice is, get started as soon as possible," says Gottbrath. "However, if it takes longer than 26 months, the optimal strategy is to delay until you can buy a computer fast enough to do the calculation in precisely 26 months."
For calculations that take longer than this, Gottbrath's model can calculate the optimal amount of "slack" time to leave before making a start. For instance, if a calculation is set to take 41.2 months to complete at current processor speeds, there is no point starting for a whole year. "We suggest going to the beach," says Gottbrath. However, after a year of "goofing about", he says you should then immediately buy a new computer and put your nose to the grindstone. "You'll still finish early, saving yourself 3.25 months of total time."
So do Gottbrath and his colleagues intend following their own advice? "Well, actually I don't have any calculations to do that would take longer than 26 months," he admits. But code breakers who have to tackle intensive problems such as finding the prime factors of very large numbers might take heed of the advice.
Gottbrath and his colleagues have posted a manuscript describing their work on the Internet, but they don't have any plans to submit it to a journal. "Basically, it's all a bit of fun," says Gottbrath.
Other researchers seem to be taking it in the same spirit. "I would be happy to comment on this work," David Hough, an astronomer at Trinity University in Texas told New Scientist. "However, would you mind if I did it next week?"
Author: Marcus Chown
New Scientist issue 8th January 2000
More at: http://agave.as.arizona.edu/ ~chrisg/mooreslaw.html
PLEASE MENTION NEW SCIENTIST AS THE SOURCE OF THIS STORY AND, IF PUBLISHING ONLINE, PLEASE CARRY A HYPERLINK TO : http://www.newscientist.com
Journal
The New Scientist