This review discusses grid computing that is a low-cost way to harness the central processing units of a group of workstations. The grid can be made up of any number of central processing units (CPU), and they may be far-flung or within the same company, or even in the same department. Grid computing puts to work on the grid all available CPUs at idle workstations and thus does away with the need for powerful servers or supercomputers. Sun Microsystems Inc., Santa Clara, CA, and IBM of Armonk, New York, have both released software within the past three years that can divide and farm out pieces of an application to several thousand linked computers. Microsoft is developing grid-computing software for use with its products, as are Hewlett-Packard, Palo Alto, CA, and others. Grid software is written in Linux, the open-standards operating system. However, because no gatekeeping technology is currently in place for grids, the hard work of IT managers seeking to implement grid technology comes when negotiating policies among departments and setting up grids accordingly.
A few years ago, Ford Motor Co.’s engine and transmission group faced the same conundrum that confronts many industries in a time of economic downturn. Ford was asking its information technology department to cut costs. But the engine and transmission group needed a lot of horsepower— or at least a lot of computer power—to run computer-intensive applications, such as finite element analysis, computational fluid dynamics, and design-of-experiments simulations.
The answer to the group’s problems lay in getting on the grid.
Grid computing is a low-cost way to harness the central processing units of a group of workstations. The grid can be made up of any number of CPUs, and they may be far-flung or within the same company, or even in the same department. One day, experts say, a huge number of personal computers might be linked in a grid to give private users access to services and much more information with quicker delivery than the Web offers today. But for now, that’s the talk of the futurists, who say grid computing contains the power to change our lives.
For now, the technology is gaining ground among large corporations, which are drawn to its low cost and its capability to direct a huge amount of joint computing power at solving a single, power-intense problem. Usually, a simulation requiring that much power is a scientific, engineering, or otherwise highly technical one.
Grid computing puts to work on the grid all available CPUs at idle workstations and thus does away with the need for powerful servers or supercomputers. That, of course, is where a large part of the cost savings comes from. A cousin to both cluster computing and parallel processing, grid computing can be thought of as distributed and large-scale cluster computing or as a form of network-distributed parallel processing. The computational grid is analogous to an electrical grid in the way power is distributed to sources along wires linked to a source of juice.
This method of harnessing computer power has traditionally been the purview of academic researchers and the national laboratories. But for the past two or three years, software and hardware vendors, always ready to tap a developing market, recognized the potential business application for this type of computing and have developed hardware and software accordingly.
The Grid Goes Commercial
Sun Microsystems Inc. of Santa Clara, Calif., and IBM of Armonk, N.Y., have both released software within the past three years that can divide and farm out pieces of an application to several thousand linked computers. Microsoft is developing grid-computing software for use with its products, as are Hewlett-Packard of Palo Alto, Calif., and others. The vendors say that one day, customers may lease computing power from hosted grids, thereby paying for only the IT power they need.
IBM’s foray into grid computing began in August 2001, but in January of this year the company launched a broad push into commercial grid technology with 10 initiatives that targeted specific business sectors: aerospace, automotive, financial, government, and life sciences. The company wanted to drive the benefits of grid computing beyond its academic and research roots and into business enterprise, according to a statement.
“This takes advantage of untapped computer power,” said Elliot Luber, a spokesman for IBM. “The average person works eight hours a day and their CPU could be used the rest of that time on a grid.”
IBM, like fellow grid provider Sun, works with middle-ware vendors to help deploy grids in businesses. In the case of the IBM aerospace and automotive grid technology, companies can set up grid platforms that encompass partner companies or suppliers in order to share data and distribute work and computational power.
For the Ford group, the answer to its cost vs. computing needs quandary came with the purchase of 500 workstations from Sun Microsystems, which group leaders planned to purchase anyway to run needed computer-aided design applications. The new workstations held dual CPUs, which got the managers thinking about a grid.
“So one CPU can stay in the grid full-time and the other is used for CAD,” said Peter Jeffcock, group marketing manager at Sun.
“And then on nights and weekends, the other CPU joins the grid because it’s not a full-time work environment at Ford. So the computers are idle.”
Engineers now send their work for analysis to the grid engine software, which automatically distributes those jobs to the machines available at the moment.
“There’s no more waiting until closing time to run power-hungry applications,” Jeffcock said. The group formerly used cluster software to make use of a computer cluster on a much smaller scale than its grid. More computer power was available to the cluster to solve problems at night, when the workstations were idle.
Right away, users began experiencing a boost in productivity because the grid software, called Grid Engine, also from Sun Microsystems, let them perform thousands more computer-intensive, what-if analyses so they could see how minute changes in a part would affect a car’s overall performance, Jeffcock said.
Engineers in the group who are performing complex design-of-experiments studies now can complete work in 15 minutes that used to take days, he added.
General Motors Corp. uses workload management software from Platform Computing of Markham, Ontario, and in-house software to tie together hundreds of workstations, to run supercomputer-class problems in aerodynamics, fluid flow in engines, and heat dispersion.
Companies like Ford and GM that need concentrated, intense computing power are the main adopters of grid technology, which Roger Germundsson calls a technology on the cusp. He is director of research and development at Wolfram Research Inc. of Champaign, Ill., which makes Mathematica software for technical computation. Wolfram recently released GridMathematica, software that is specifically formulated to break down large, mainly mathematical problems and farm them out across a computer grid.
The grid technology is particularly useful for analysis of uncontrollable or unknowable factors, like designing to account for the wind shear around a car.
“You have to test for all these controls and that necessitates an immense number of simulations,” Germundsson said. “It’s cheap and cost effective to farm them across a grid.”
Late last year, researchers in the NASA Advanced Supercomputing Division automated the CFD process that engineers go through when running analyses to predict fluid flows during aircraft design. The automated CFD process, called AeroDB, runs across NASA’s Information Power Grid, a geographically distributed network of computing and data resources.
Though CFD has helped greatly reduce the time and expense involved in designing aircraft and spacecraft, it’s still labor intensive, according to the NASA division’s publication, Grid-points. By automating it, engineers didn’t have to program each analysis separately into the computer.
“The AeroDB system significantly simplifies the process of executing many CFD jobs,” said Stuart Rogers, the senior scientist on the AeroDB project. “It’s a big step toward automating the process to take the user out of the loop of having to monitor every single job.”
The process of setting up flow-parameter studies by engineers for each design is also error-prone. By reducing dependency on a user and automating that process, AeroDB decreases the amount of error in analysis, Rogers said.
Sharing the Power
Though already in place at large-scale organizations like NASA, grid computing might face adaptation problems at some large companies because, unlike cluster computing, a grid frequently needs to stretch across departments or partner companies. Of course, politics arises inevitably.
“Any one person can use the entire set of resources available to the grid,” Rogers said. “So one engineer can send off 50 simulations at the same time and clog everything up. You’ve got to have mechanisms in place to make sure a project doesn’t do that.”
Because no gatekeeping technology is currently in place for grids, the hard work of IT managers seeking to implement grid technology comes when negotiating policies among departments and setting up grids accordingly, Jeffcock said.
“It’s hard work,” he admitted. “But the benefit is, as business needs change—one project might need different computing resources than a past project, for example—resource allocations can change. If I’ve been allocated 30 percent of the grid in the past and for a future project I need 40 to 50 percent of the grid, IT could make those changes with no problem. Discussing among various vice presidents why you should make that change might take longer than reconfiguring the software.”
The future might more commonly hold grids shared among business partners, although such a thing is rare today, Jeffcock admitted.
“The fact is, the majority of grids are surrounded by a firewall for good reason: security,” he said. “Would an organization be comfortable sending proprietary design information across a network? No. We hope in the future standards and technology will make that easier to do.”
Getting the Grid in Synch
Grid software is written in Linux, the open-standards operating system. Standard methods of constructing and maintaining grids so they can be uniformly adopted and executed—like gatekeeping ability—are still being written and adopted for this fledgling technology.
Most of the corporate providers of grid technology have signed on to the Globus Project, a research and development project started in 1996. It is focused on developing standards for grid computing so that all software and hardware works in a methodical and predictable way, and can work together. The project is centered at Argonne National Laboratory’s Mathematics and Computer Science Division, the University of Southern California’s Information Sciences Institute, and the University of Chicago’s Distributed Systems Laboratory. It’s a conglomerate of academic, commercial, and corporate partners. Work includes setting security and authentication standards such as exist currently for the Internet.
But grid computing is useful for more than solving large-scale scientific and engineering problems, according to the folks at Globus.
Imagine the types of applications that might be constructed if access to supercomputers, live satellite imagery, and mass storage were as straightforward as access to the Web, they say.
Last summer, the National Science Foundation began to install its TeraGrid, a transcontinental supercomputer that should do for computing power what the Internet did for documents. First, clusters of high-end microcomputers are being set up at four sites: the National Center for Supercomputing Applications at the University of Illinois at Champaign; the U.S. Department of Energy’s Argonne National Laboratory outside Chicago; the California Institute of Technology in Pasadena, and the University of California’s San Diego Supercomputer Center. This year, these four clusters will be networked to behave as a single entity, providing a look at the potential future of grid computing.
When it is up and running, the TeraGrid will be able to solve problems eight times faster than any academic supercomputer that is in existence today.
With that kind of speed, scientists can tackle some of the computationally intensive tasks they face—problems like protein folding that will form the basis for new drug designs, climate modeling, and deducing the content and behavior of the cosmos.
In the 1980s, the National Science Foundation created the NSFnet, a communications network that gave scientific researchers easy access to its new supercomputer centers. It was eventually picked up by the business sector, then widely adopted by private citizens. NSFnet did change life as many know it. Now it’s better known as the parent of the Internet.
Indeed, many say that grid computing is poised to become as big in the future as the Internet is today, in its ability to link everyday users to all kinds of information as well as to perform high-end calculations that are hard to run on even the largest supercomputers.