Analysis: Grid ComputingBy Bill Roberts | Posted 12-01-2001
Analysis: Grid Computing
Click image to download full Grid Computing chart in PDF format.
Back in the 1960s, when computers were rare and expensive, CPU time was scheduled and billed by the millisecond. Now, CPU time is cheap, so much so that computers these days spend most of their time displaying your vacation photos of Yosemite or simulating an aquarium. These simple "screen saver" programs, which compute nothing and whose only purpose is to stir up pixels on the display screen, probably consume more of the world's computational capacity than any other kind of software. Go into almost any office and you'll find machines busily saving their screens all night and all weekend.
But even when a computer is in use, it's mostly idle and operating way below peak powers. The average office worker might produce 10 keystrokes a second; that's not much of a distraction for a processor that can execute 100 million instructions in that same blink of an eye.
A lot of companies might have a great need for some of that unused computing powerand not just that being squandered by the PCs in your office. Imagine if all the PCs in Manhattan, for exampleor the U.K., for that mattercould be focused on solving your company's latest global problem or on discovering a new source of oil or a new drug to cure arthritis.
GlaxoSmithKline PLC is already thinking in such terms. Like every pharmaceutical company, GSK faces a high-pressure challenge: To discoverfastthe few molecular compounds that can be used to make new drugs. To do that, tens of thousands of compounds must be screened, mixed and matched in order to find the few dozen suitable for development. That's no easy challenge: Drug development is a marathon that lasts years, so if makers can cut even one week from the time it takes to get a new drug to market, it can save millions of dollars. If GSK could boost computing power enough to significantly hasten the drug discovery process, it could beat more rivals in the race to market new drugs. "Ninety percent of drug candidates fail to make it to market," says GSK grid strategist Joseph Simpkins. "So it is important to fail quickly."
Simpkins, therefore, is always on the prowl for more computing power. In the past, GSK ran its custom codethe algorithms that determine if a single compound has drug potentialon several dedicated UNIX workstations, and it could take weeks, if not months, to get results. Three years ago, Simpkins redoubled his efforts to find a faster solution. His first instinct was to buy a multimillion-dollar supercomputer, but that would have been too expensive. Simpkins also toyed with customizing computer chips to run the code fasterbut that would have required reprogramming the supercomputer or creating new custom chips each time a new algorithm was developed. Both options were too costly and inflexible.
Then Simpkins had a hunch. Like most companies, he figured, GSK didn't use all the computing cycles on its desktop PCs. To prove his theory, he surveyed a sample of the several hundred PCs in the research division and was astounded by the results. "The unused computing cycles were close to 95 percent at any time of day." Simpkins had found his solutioncomputing power available right under his nose. GSK scientists now tap unused power over the corporate intranet from GSK's 600 desktop PCs in Britain, Italy and Research Triangle Park, N.C. At peak times, GSK's in-house grid uses 95 percent of the company's total PC capacity. "Our vision is to harness all available desktop power in the company," says Simpkins. "This has opened up a whole new way of thinking, and the scientists started to come up with new ways of doing their research that they hadn't thought of before."
GSK isn't alone in contemplating the possibilities. Other companies, from The Boeing Co. and Ericsson Inc. to Unilever PLC, are also experimenting with in-house grids, and the payoffs on collaboration and data-sharing can be significant. At Boeing, a home-styled gridwhich includes several large supercomputers and many desktop machines at eight separate Boeing sitesrecently helped engineers solve a design problem in seven days, mustering 96,000 hours of computing cycles during that period. "This computation would not have been attempted before, as it would have taken about seven years to do on a contemporary high-performance workstation," says Kenneth Neves, director of research and development planning and collaboration for Boeing.
Today, GSK's three-year-old grid project has started to see a payoff. It has already negated the need to spend $4.8 million on 300 dedicated UNIX servers that would have been needed to achieve higher-speed new-drug screening. And GSK grid specialist Simpkins says grid computing can cut the time it takes to analyze a single compound from eight weeks to as few as two, cutting R&D costswhich for GSK, Wall Street analysts estimate, amounted to some $10 million a day last yearat the same time. But to Simpkins, the most important payoff is the competitive advantage GSK will get from being faster at analyzing thousands more compounds. "The potential return on investment is mind-boggling," says Simpkins. GSK CIO Ford Calhoun is similarly optimistic, and recently set up a grid task force of IT and R&D staff to cull a list of specific business projects that could benefit from a surge in computer power.
But these in-house, client-centric experiments in distributed computing could soon be rendered child's play. In recent months, a number of companies have begun the early spadework needed to build a nationwide computing grid that would, some scientists envision, be analogous to today's electricity grid, putting one company'sor hundreds of companies'computing power, distributed via the Net, into the hands of other companies that need it for everything from drug research to collaborative automotive design to high-speed consumer data analysis. Often called the next step in the evolution of the Internet, business grid computing will allow users to share computing power, data and applications as easily as information is exchanged today over the Web. Grids create vast pools of computing power by connecting computers from disparate locations using the Internet or high-speed networks as well as open-source protocols (see
A whole new industry is already springing up around grids. In August, IBM put its full weight behind Globus, an international consortium founded in 1996 that is developing an array of open-source software that will enable companies to securely get their systems onto an open, cross-company, cross-industry grid. Since then, dozens more companies, including Sun Microsystems Inc. and Japan's NEC, have come forward to help develop the new grid.
IBM's support of efforts to develop grid computing on open-source software has been an important catalyst for the commercial development of the technology. Turek says IBM is in discussions with companies in five sectorsfinancial services, automotive, aerospace, pharmaceuticals and energyto help them either build or run grids aimed at boosting their ability to collaborate in-house across geographical boundaries, get products to market faster and crunch numbers quicker on complex problems. In November, IBM announced plans to help build the grid technology to support a consortium of more than 60 life-sciences companies and research organizations. Grid computing is "gaining critical mass" among those in the business community, says grid expert Ian Foster, head of Globus, chief of the distributed computing systems laboratory at Argonne National Laboratory near Chicago and a computer science professor at the University of Chicago.
IBM's Turek says the commercial opportunities are promising. Some companies, he says, might see a chance to create markets for CPU cycles or data cycles. Others, like IBM, might build businesses around offering companies a way to build their own grids that, using open-source software, would offer access to a common, utility-based grid. Still others predict that new companies will be created to Web-host the new global computing grid in the same way that utilities now exist to provide electricity to homeowners and businesses. "When you plug your appliance into the wall, you don't know who's supplying you with electricity that day. There may be a lot of generators, but all you know is that you get a bill from your power company for what you use. You don't really care which generating facility produced the electrons you're making use of that precise second, but you don't have to. What you care about is that when you hit the switch, you get the power."
In all, Turek predicts the fledgling new grid computing industry could grow to be worth billions. "It's not going to be limited to just millions of dollars of opportunity. We believe this will grow to billions of dollars, not just for us but for the industry at large," he says.
Grid computing isn't entirely new. It's actually a variation on the notion of distributed computing, which takes an application off one server and runs it more effectively by spreading it across several servers on a network.
A few early experiments with distributed computing, including a pair of programs called Creeper and Reaper, ran on the ARPAnet, the 1970s predecessor of today's Internet. Later, when Xerox Corp.'s Palo Alto Research Center (PARC) installed the first Ethernet, a program cruised the network at night, commandeering idle computers for CPU-intensive tasks. A later scavenger system, called Condor, developed by Miron Livny and his colleagues at the University of Wisconsin at Madison, is now running at more than a dozen universities and other sites. Condor (see "Flight of the Condor" in this article) roams within clusters of UNIX workstations, usually confined to a single laboratory or department, and delivers 400 CPU days per day of free computing to academics at the university and elsewheremore than many supercomputer centers.
In 1995, grid computing concepts were explored by computer scientists around the country in a project organized by the National Science Foundation called I-WAY, in which high-speed networks were used to connect, for a short time, high-end resources at 17 sites across North America. Out of this activity grew a number of Internet grid research projects. In 1997, the Entropia network, spearheaded by Argonne's Foster, was established to apply idle computers worldwide to problems of scientific interest. In just two years, this network grew to encompass 30,000 computers with an aggregate speed of over one teraflop per second, or the capacity to perform 1 trillion numerical calculations per second. Among its scientific achievements is the identification of the largest known prime number (see www.mersenne.org). Since then, venture funding has been raised to use grid computing for philanthropic concerns, including Parabon Computation Inc.'s Compute Against Cancer project, which analyzes patient responses to chemotherapy, and Entropia Inc.'s Fight AIDS at Home project, which evaluates prospective targets for drug discovery.
Now, grid computing is "gaining critical mass" in the business community, says Foster. Indeed, hardly an industry exists that doesn't need to crunch more data now than it did 10 years ago. Modeling and simulation software are available for everything from airplane wing design to data mining. This comes as CIOs feel enormous pressure to squeeze the most from their budgets. At the same time, their bosses demand they find new ways to use technology for competitive advantage. The grid strategy can help.
The right circumstances include research, engineering, financial analysis and other areas that need heavy number-crunching, usually on smaller sets of data that require hundreds or thousands of calculations. At pharmaceutical firms, that means drug discovery. At Oracle Corp., grid computing is used to run more than 50,000 tests daily on the database software it develops. And Motorola Inc. uses grids in many areas, including verification tests on code for the semiconductors it builds. Intel Corp., meanwhile, says its internal grid has saved it $500 million over the past decade.
Just the Beginning
Just the Beginning
It will be some time, though, before grid computing is ready for prime time. Databases aren't designed for distributed computing or the grid model yet, so the grid can't yet run corporate financials or other applications heavily dependent on databases. "When you have ad hoc computing where it is hard to predict demand day in and day out, the grid is very attractive," says Jack Cooper, CIO at Bristol-Myers Squibb Co., a New York-based drug company that is pilot-testing the grid strategy. "But it is not for standard transaction processing."
But IBM's Turek disagrees. So far, he concedes, "the bulk of the applications to which grid computing has been applied have been driven by people in the university research environment who are certainly more biased toward the computational. But everybody who works in this area knows that the management and analysis of data is the end game. How do you resolve grid approaches to both transactional and batch-oriented data analysis? It's just a matter of advancing the technology a bit from where it is today."
Forward-looking CIOs are already working closely with business leaders at their firms to consider the potential that both the client-centric, in-house grids and the coming utility-based grids might hold for the future business strategies of their corporations, says Foster. "As the business world's understanding of grid strategy broadens, savvy CIOs and their staffs will likely find new tactical and strategic ways to use it," he says. Foster, for one, says it will be no time at all before someone will try to turn excess PC capacity into a paying proposition. "Huge outsourcing service provider models are just around the corner," he says. The Burton Group's Mike Neuenschwander says CIOs and their staffs should think about the possible applications for grid computing in their own companies. "If CIOs think about networking and harnessing these underused computing resources, then they can recognize the business possibilities for grids in many places," he says.
GSK's Simpkins is already mulling the business possibilities. Today, GSK only focuses on its own internal needs. Tomorrow, Simpkins believes, the company will have the ability to sell unused computing power to outsiders if it chooses. "The grid could be a new revenue center," he muses. Not a bad idea for a company already in the business of discovering new cures for the bottom line.
BILL ROBERTS is a Silicon Valley-based writer, editor and technology journalist whose work appears regularly in a variety of technology publications, including eWEEK and Internet World. Comments on this story can be sent to firstname.lastname@example.org.
Flight of the Condor
Flight of the Condor
The patterns of use in a grid computing array, taken from the usage log for one day of the Condor grid project at the University of Wisconsin. Condor regularly delivers 400 CPU days per day of essentially "free" computing to academics at the university and elsewhere, more than many supercomputer centers.
At any moment during the day, on average...
- 14.2% of computers were in use by their owners
- 78.9% were in use by Condor
- 7% were idle
- 25% were in use by owners (peak time)
- 89% were in use by Condor (peak time)
SOURCE: THE CONDOR PROJECT/UNIVERSITY OF WISCONSIN-MADISON
The National Science Foundation's supercomputing grid, to be completed in 2003, will be able to perform 11.6 trillion calculations per second and store more than 450 trillion bytes of data over a 40-billion-bits-per-second optical network.
SOURCE: NATIONAL SCIENCE FOUNDATION
The Grid: Blueprint for a New Computing Infrastructure
By Ian Foster and Carl Kesselman.
Morgan Kaufmann Publishers, 1998
A magazine about grid computing and related topics.
The Globus Project
The Global Grid Forum
The National Virtual Laboratory for Earthquake Engineering
Grid Physics Network
The search for alien life in outer space
Compute Against Cancer