Vast Reservoir of Computing Goes Unused
Every minute of every day, a foolish and wasteful American population casually destroys the computational equivalent of roughly 1 million Cray YMP supercomputers. This is technological waste of epic proportions. You can help prevent this wanton squandering of our precious digital resources--or you can turn the page.
Even as you read this, millions of personal computers and workstations are sitting on desks across America, silently waiting for something to do. Won’t you help them?
According to surveys conducted by the Lawrence Livermore National Laboratory and others, these machines lie idle 95% to 97% of the time. For more than 22 hours a day, these high-powered microprocessors aren’t solving any problems or adding any value; they’re merely wasting computational cycles.
Multiply this waste by the millions and you’ll know we’re burning a Brazilian rain forest of computing every day. (Of course, on the scale of political correctness, preserving computational cycles isn’t nearly as fashionable as preserving air, water or endangered species.)
Consequently, comrades, there is a specter haunting computerdom--the specter of computational communism: From each processor according to its abilities; to each according to its needs. Why not tap into this vast reservoir of unused computing? Stop treating personal computers and workstations as mere individuals or networkers; start treating them as part of a giant metasupercomputer that is fuel-injected by recycling all these wasted processors.
“We’re powering several critical computational applications for our physics and chemistry departments literally out of the recycled garbage of the computer science department,” says David Gelernter, a Yale University computer science professor. In other words, Yale’s science computations are being powered by all those idle processors.
Gelernter has been a pioneer in promoting what he calls “ensemble computing”--or, more colorfully, “piranha processing.” In effect, use all these multiple processors to nibble the problem to death.
The logic is childishly simple: If two computers can solve a problem twice as fast, 50 computers can solve it 50 times as fast. It turns out that most complex problems lend themselves to this sort of “parallel processing” solution.
It’s the reverse of computer time-sharing; instead of users drinking from a central reservoir, each user can steal a sip from any machine on the network. By connecting through a special operating system, individual computers are thus melded into a virtual supercomputer. Even as part of your PC munches on your word processor, it can also be nibbling on a complex engineering calculation. Problems too tough for your PC are automatically cracked into digestible chunks and sent to other processors.
As a rule of thumb, Gelernter estimates that 30 to 50 Sun Microsystems workstations roughly equals one Cray. Indeed, Sun Microsystems has a program called Godzilla that turns Sun networks into parallel processing supercomputers. As the microprocessors become more powerful, so will the virtual supercomputers.
Of course, some people may find sharing their computer as distasteful as sharing their toothbrush or lending their car keys to an accident-prone colleague. The trick will be to create metacomputer operating systems that make the actual operation of the virtual supercomputer invisible to individual users.
However, the idea of “buying” a supercomputer that’s already there is incredibly alluring to capitalistic companies looking to get more for less. If a company can recycle paper, why not recycle computer cycles? Too many companies simply buy another machine rather than effectively reuse their existing computer “waste product.”
“This new paradigm is the crucial sociological foot in the door,” Gelernter asserts, “because companies aren’t interested in investing in exotic supercomputing hardware. They’d rather invest in what they already have. (What’s more) because of the cost-effectiveness of the approach, top management becomes interested. . . . You’ve invested in all this equipment that’s doing nothing most of the time; why not acquire this reservoir of unused power?”
Indeed, cutting-edge companies such as Motorola have just begun to explore the idea of using software to transform their installed base of workstations into metasupercomputers. Wall Street firms--which make as heavy use of computers as virtually any American company--are also investing internally.
It’s too early to determine if digitally knitting PCs and workstations together into metasupercomputers poses a competitive threat to genuine supercomputer companies such as Cray, Convex and Thinking Machines. But the odds are this trend toward computational communism will make their economic advantage more difficult to discern.
To be sure, some people may argue that since computing power is becoming cheaper by the minute, who cares if we waste a few cycles here and there? But then, isn’t that what they used to say about the ocean and the atmosphere? Remember, a microprocessor is a terrible thing to waste!
