Imagine if computing power were as easily accessible as electricity.
Businesses essentially would be able to tap into a vast network of supercomputers and pay only for the power they use, rather than spend millions of dollars on the hardware itself.
Such a scenario could become a reality in a few years as international efforts to develop computing grids heat up and major computer makers such as IBM seek to develop powerful networks that companies can tap into as needed.
In the last month, major initiatives to create powerful grids linking thousands of computers have been announced in Great Britain and the United States.
Two weeks ago, the National Science Foundation contracted with Intel Corp., IBM and Qwest Communications International Inc. to build a $53 million grid. Once completed in 2003, the grid will link computers at four research centers and will be able to perform 11.6 teraflops (trillion floating-point operations per second) and store more than 450 trillion bytes of data over a 40 billion-bit-per-second optical network, according to the NSF.
A week earlier, IBM, of Armonk, N.Y., announced it had been chosen to help build the United Kingdom National Grid and disclosed plans to build its own grid by using 50 computing “farms” around the world.
While todays largest computer grids are used primarily for scientific and research purposes—such as mapping human genes—the development of standard grid protocols could lead to commercial grids.
Under such a scenario, a company in need of more computing power could tap into the grid and gain access to a teraflop of CPU power or more, paying only for what it uses.
But such commercial computer grids likely are still several years away as research has yet to overcome such obstacles as security and bandwidth concerns.
However, ongoing research is benefiting businesses in other ways by enabling computer and software companies to develop new technologies, said Andy Butler, an analyst with Gartner Group Inc., in Egham, England.
“Computer grids are a very good testbed for leveraging architecture designs that will probably then feed back into more modestly scaling commercial applications,” Butler said.
Already, large corporations such as processor makers are creating internal grids to handle complex computer simulations. “Within Sun Microsystems [Inc.], we have a compute farm that totals about 3,000 CPUs, which we use in the course of designing the microprocessors that we build here,” said Peter Jeffcock, group marketing manager in Suns Volume Systems Products Group, in Menlo Park, Calif. “We leverage that power to run different simulations. For example, we can test to see what will happen to a chip if we supply it with 5.1 volts or 5.2 volts, as well as what will happen if we do all that at different temperatures.”
“In most companies, you typically find that workstation CPUs are only being used 5 [percent] to 10 percent of the time in a 24-hour day,” Jeffcock said. “If you integrate them into a local computer farm, you can get system use in the 80 [percent] to 90 percent level.”
In a move that could persuade even more companies to adopt grid computing, Sun last month released its Grid Engine software to the open-source community.
The software essentially locates idle computer resources, matches them to individual job requirements and delivers networkwide computing power to the desktop, effectively managing an organizations computing resources and job distribution.
The free Grid Engine software has proved a valuable resource to Matt Ferris, systems manager with Motorola Inc.s Semiconductor Products Sector, in Tempe, Ariz.
“Weve been using it since February to help us handle test and verification jobs on wireless and broadband circuit designs,” Ferris said. By using the software, he said, Motorolas engineers can submit 40 to 50 jobs into a queue, where the software will automatically distribute the workload to various computer systems as they become available, including overnight when workers have gone home.
“Before, we had to do it by hand, hunting around for machines where people werent logged in,” Ferris said. “Now they just put it in a queue; they come in the next morning and review the results.”
Because of the enormous potential to offer users access to vast amounts of computing power, its only a matter of time before grids become commonplace in corporate environments, said Suns Jeffcock. “We think of this as being something that is inevitable,” he said. “The gains are so big, and the benefits to the companies are so huge, that this is an inevitable transition. Its just a matter of when and how fast.”
John Patrick, IBMs vice president for Internet strategies, agreed. “Ive often been asked, Whats the next big thing for the Internet?” Patrick said. “Until now, I didnt have the answer. Im very confident now that the next big thing will be grid computing.”