Plotting A Grid Strategy

Some pioneers are already using grids in data- and CPU-hungry pursuits such as life sciences R&D and engineering and design, but experts say widespread business use is far away.

Until recently the purview of astrophysicists and drug designers, grid computing is a prime example of "latest-greatest syndrome" in the enterprise technology business. Some bleeding-edge companies are already using grids in data- and CPU-hungry pursuits such as media production, life sciences research and development, and engineering and design, but experts say widespread business use is still several years off and not without caveats.

Nonetheless, computer industry heavyweights and assorted start-ups are staking out turf in the brave new world of computational grids—potentially vast agglomerations of computing muscle and data that promise to fuel e-commerce and enterprise applications while also serving scientific and technical needs.

Grid computing emerged as a means for scientific researchers to tackle complex problems, the calculation and processing of which would overwhelm even the most powerful supercomputers. Whether they scrape together CPU cycles from idle workstations or harness multiple supercomputers, grids provide the framework for assembling sophisticated, ultrapowerful virtual computers.

These assemblages of distributed resources are coordinated by software that handles the differences among computer operating systems and manages things like scheduling and security. Grids can make available a wide range of CPU cycles, storage and other resources.

Grid Alert

Grids have generally been used in business for some time in the form of search engine farms and systems handling bioinformatics and media special effects, among other applications, notes Nick Gall, an analyst at Meta Group Inc. in Stamford, Conn.

The grid rubric covers a wide range of private and public systems, some under the control of single entities and others run by coalitions. To date, many grids have taken shape in the academic and government research communities, some on a nationwide scale. Others are internal corporate setups.

Some of the more high-profile grids include government projects such as the National Science Foundation's National Technology Grid, NASA's Information Power Grid and the European Union's DataGrid. There are also private corporate efforts in the aerospace, pharmaceutical and automotive industries.

Grids could set the stage for a more adaptable and viable second wave of application service providers (ASP) and their acronymic cousins, xSPs. They could also form the basis of more flexible and powerful intranets and extranets.

Part of the attraction is the prospect of putting the right mix of CPU power, data and bandwidth at users' disposal as they need it, whether that's supplied through a private, highly secure grid configuration or a loosely coupled publicly available grid.

The prospect of plugging into grids instead of laying out huge sums of money to upgrade internal systems has an obvious appeal to companies trying to weather a down market.

"[Given] the paucity of skilled people out there and the cost of training, there will be a progressive desire to [create a] coagulation of technology in a way to really leverage the skills that do exist and still provide the level of service users are really looking for," says Dave Turek, vice president of emerging technologies at IBM.

In addition to technology heavyweights such as IBM, Sun Microsystems Inc. and Hewlett-Packard Co., smaller players such as Markham, Ontario-based Platform Computing Corp., San Diego-based Entropia Inc. and Austin, Texas-based United Devices Inc. are pushing various enterprise and vertical-industry grid tools as well.

But don't look for grids to overlay the enterprise for at least a few more years, because the infrastructure tools and business adaptations are still taking shape, according to observers.

"Full-blown grid computing is still about two to three years out, and five years from being a big impact," says Gall.

To that end, the business models for delivering, supporting and charging for grid services and applications also have yet to jell. And while it might be tempting to view grids as the one hammer for all your nails, the trick for IT managers is to identify if and where grids are appropriate. Work remains on such key functions as load balancing, cluster management, resource identification and sharing and database integration, as well as ensuring a useful degree of standardization, according to observers.

Gridlock

Indeed, the transition won't be automatic, since R&D labs have different computing needs than corporate IT centers.

"A grid as an object that scavenges spare cycles from someplace is less useful to the enterprise than a grid as a mechanism for locating data resources," says Robert Hollebeek, a physics professor and director of the National Scalable Cluster Project at the University of Pennsylvania in Philadelphia.

As a first step, IT managers would be wise to ask, "Are they really in an enterprise that could benefit from better communication and sharing of resources in a wide-area networked setting? If so, they could start adopting some authentication mechanisms so they can participate in a grid with trusted entities," Hollebeek advises.

"In the context where it's hard to predict how much you'll need to ramp up your systems and what you'll need to integrate, grid computing provides some flexibility," says Paul Kearney, director of bioinformatics at Caprion Pharmaceuticals Inc. in Montreal. "It's not clear to me that someone would choose to migrate to grid computing when their data is more homogeneous and integration of technology moves at a slower rate."

Plus, computers strung together in a distributed fashion would be virtually useless for many low-latency applications, which would be bogged down by the Internet's communications delays.

"I don't find the concept of lashing computers together with grid middleware to do very large calculations appealing. Many of the compute-intensive applications of interest to us have stringent latency and bandwidth demands," says Thom Dunning, director of the North Carolina Supercomputing Center in Research Triangle Park and vice president of high-performance computing and communications for its parent organization, MCNC. CPU-hungry enterprise applications fall into this category.

Finding a Fit

The appropriateness of grid computing depends on whether businesses "have problems that need significant computational resources and can be distributed," says Alex Bangs, chief technology officer at Entelos Inc., a Menlo Park, Calif.-based bioinformatics company. "Do they really need to sit on some giant 64-processor machine or some big mainframe, or can they be broken up into small pieces and those small pieces be crunched on a smaller machine, whether that machine is sitting in a server room somewhere or whether that machine is some idle cycles from somebody's desktop?"

Entelos develops complex computer models of human diseases and virtual laboratory software for researching the diseases and developing treatments. The company works with pharmaceutical giants such as Aventis, Pfizer Inc. and AstraZeneca PLC, whose vast intranets and countless desktops make them ideal grid computing users. Both Entelos and the pharmaceutical companies use grid computing to fuel the heavy-duty processing needed to run the virtual experiments.

"We're doing a specialized kind of modeling. In general, people do all kinds of models—there's financial models, and even the data mining is a lot of times based on a model," Bangs says. "What scaling up to grid computing means is being able to look at many, many more variations of those models."

Players in other industries such as automotive design are also beginning to harness the potential of grid computing.

"We're working on the complete aerodynamics of the race car," says Kevin Colburn, team leader for computational fluid dynamics at the West McLaren Mercedes Formula One racing team in Woking, U.K. "We're concentrating on typical kinds of aerospace things like the front and rear wings of the car. But we also analyze suspension components for their angle and orientation, [as well as] radiator flow, air box, you name it."

More important, grids "allowed us to do the same amount of work quicker, and then allowed us to research other areas of the car that we may not have done previously because we didn't have the compute capability," Colburn says.

Bowen is a contributing editor at Technology Research News. Contact him at ted_bowen@hotmail.com.

Copyright © 2002 IDG Communications, Inc.

  
Shop Tech Products at Amazon