Slow Going On The Global Grid

Feb 20, 2005 (07:02 PM EST)

Read the Original Article at

When grid computing leaped from the nation's research labs into business computing three years ago, its promise of delivering supercomputer power to desktop PCs and unshackling users from the limits of their department's technology looked boundless. Grid computing, powered by public-domain software called the Globus Toolkit that was spreading across university and government labs, would pan the Internet for the computers and databases best able to solve complex calculations. Tech-industry CEOs, including IBM's Sam Palmisano and Hewlett-Packard's recently deposed Carly Fiorina, pumped up the benefits of data centers that could configure themselves to grab computing power from in-house or faraway machines. Microsoft chairman Bill Gates told InformationWeek three years ago that grid computing represented "the Holy Grail of computer science."

But a funny thing happened to grid computing on its way up the corporate ladder. Practically no one in business is running the current Globus software that HP, IBM, Microsoft, and others were counting on to organize their customers' computers, disk drives, databases, and operating systems into powerful networks that would span organizational boundaries.

Johnson & Johnson is no longer sponsoring the Globus Toolkit, but grid computing is saving its R&D unit money, David Neilson, Johnson & Johnson's Dir. of Drug Discovery says. -- Photograph by Ken Schles

Johnson & Johnson is no longer sponsoring the Globus Toolkit, but grid computing is saving its R&D unit money, Neilson says.

Photograph by Ken Schles
Johnson & Johnson's $22.1 billion pharmaceuticals research-and-development unit has two grid projects under way, each geographically contained and centered on its own networks. David Neilson, director of drug-discovery information management, expects these projects will save the company "in the seven- to eight-figure bracket" over five years by recruiting low-cost computers to handle tasks once done mostly by expensive specialty machines.

Johnson & Johnson was an early sponsor of the Globus Toolkit project, an effort by researchers at Argonne National Lab, the University of Chicago, and the University of Southern California to develop open-source software that could link distant computers and users over the Net. "We're not [a sponsor] anymore," Neilson says. For its virtual drug-screening pilot in Belgium that recruits the spare power of 400 PCs and 64 Linux servers and for a stateside effort to model clinical drug trials using more than 100 PCs and workstations, Johnson & Johnson has been using commercial software from United Devices Inc., which can schedule jobs across PCs or servers under one roof, but whose capabilities fall far short of the promise of building global grids.

So why isn't Globus getting a chance? "I'm intellectually interested in it," Neilson says. "But for something that has to work 12 hours a day, we're much more inclined to go with a product that can be brought in quickly" and includes tech support to "protect the capital investment."

An ever-changing technology road map and a current version that's considered balky and out of step with industry standards not only has created a problem for Globus, but it's holding back grid computing from fulfilling its original promise. Sun Microsystems, United Devices, and a handful of small companies supply software that can distribute computing tasks across in-house machines, making sure their power gets maximized. But only Globus can create virtual teams of users at different companies, government agencies, and universities. Far-flung users working with NASA or Europe's CERN particle-physics lab can sign onto a grid of machines running Globus, and the toolkit automatically figures out what IT they're allowed to access, and when.

The perception that Globus is fine for eggheads but not a solid product for business could be about to change. This week, the researchers who wrote it plan to issue a second test release of version 4 of the Globus Toolkit, which should improve spotty quality and at long last meld the grid-computing protocols that let users write to a common software interface to share CPU cycles, files, and data with the Web-services standards. A final version of the software is due in April from the Globus Alliance, the open-source project founded by Ian Foster, a senior scientist at Argonne and a University of Chicago professor; professor Carl Kesselman from USC; and Steve Tuecke, a former Argonne manager.

In December, the three spun off Univa Corp., a private company, to take on the development, licensing, bug-fixing, and tech-support chores that could make grid computing, Globus-style, more attractive to companies. In addition to the business-friendly version it plans to sell, Univa will continue to offer a free version under the software's open-source license. Last month, they also got HP, IBM, Intel, and Sun to put up $250,000 each to fund the Globus Consortium, which plans to port Globus to Windows (it currently runs on Unix and Linux); integrate it with enterprise-resource-planning applications, clustering software, and identity-management systems; manage the software's technical road map; and help fix bugs in a predictable way.

"For [global grid computing] to take off commercially, it was going to require us to do it," says Tuecke, Univa's CEO. Globus can do technical computing out of the box, he says, but it wasn't built to handle business apps.

Globus founder Ian Foster gears up for new release.

Globus founder Ian Foster gears up for new release.
Globus version 4 includes a software layer, the Web Services Resource Framework, that fuses Web-services standards for functions such as security and messaging with grid-computing protocols, so users can write software to a common interface that runs on a grid without knowing the technical ins and outs of each machine it touches. "The mismatch between grid services and Web services is going away," says Greg Astfalk, chief scientist in HP's office of strategy and technology.

That's good news for big tech vendors that think Globus has the potential to create demand for their products. According to research firm IDC, worldwide sales of software for virtual processing, which includes grid computing, grew 22% last year, to $1.5 billion, compared with just 6% for operating systems. SAP is testing special grid versions based on the Globus Toolkit of three of its business apps--E-commerce, manufacturing planning, and human resources--with a handful of U.S. and European customers. "They love it," says advanced technology VP Vishal Sikka. "The big benefit would be not having to plan how this software is run. That's a major [total-cost-of-ownership] saver." SAP plans to include grid-computing functions in the next version of its NetWeaver development and run-time environment, due next year. Sun is looking at building products based on GT4. And at the GlobusWorld 2005 conference in Boston this month, IBM said it will include the Web Services Resource Framework in the WebSphere Emerging Technology Toolkit, now being tested, and the next full release will include the technology. Lining up Globus with Web services should "spark the development" of new apps and expand the market," IBM VP Al Bunshaft says in an E-mail.

It has been a long road to get to this point. Few vendors have aggressively packaged the current Globus version, now nearly 3 years old, with their products. Programmers lacked standard APIs for getting data into and out of grids, an improvement promised in GT4. A set of specs called the Open Grid Services Infrastructure for providing data security, balancing workloads, and transferring files wasn't compatible with Web services, and it didn't work well with widely used development tools. "The commercial tools out of the box would barf on this," says Marty Humphrey, a computer-science professor at the University of Virginia.

Even with the improvements, IT uptake may be slow. GlaxoSmithKline plc, the $38 billion-a-year drugmaker, has been using grid-computing technology from United Devices to build a "pretty powerful virtual supercomputer" from 1,500 desktop PCs in the United States and England and to schedule jobs on clusters, says Sam Thomsen, president of North Coast Idea Co., a Glaxo consultant. Thomsen is part of a committee that provides feedback on Globus to Foster and his collaborators, and he says GT4 is a "major contender" for use in an EU-funded grid project to encourage companies to collaborate in the auto, aerospace, and pharmaceutical industries, which Glaxo has joined. But Globus' emphasis on creating virtual teams runs against drug companies' reluctance to share anything of intellectual value. "That's becoming more interesting, but there's still not a lot of collaboration in the pharmaceuticals business," he says. "Globus almost adds a layer that's completely unnecessary."

Philosophical problems aren't the only ones dogging Globus. For one thing, its founders aren't even sure who's using the software, so it's hard to gauge design requirements. Foster counts 50,000 downloads from Globus' Web site last year, though "that's not a very useful piece of data," he admits. IBM, for instance, won't say which companies or how many are using its Grid Toolbox development tools, which incorporate Globus. Maintenance costs also are too high. Installing and managing grid software eats up about a third of grid projects' IT budgets, according to a survey of more than 60 companies by IT research outfit the 451 Group.

Nor has Globus retained support from some big tech vendors. Absent from the Globus Consortium are Microsoft, whose grid plans amount to one big question mark, and Oracle, which trots out the word "grid" every time it talks about clustering its databases but hasn't built Globus compatibility into its products. "We need to see whether this thing's stabilized enough to ship it to our enterprise customers," says Bob Thome, an Oracle senior distributed database manager. Microsoft's research department granted $1 million to the Globus Project three years ago to try to find a way to port the software to Windows. But that project is considered dead. "I don't need the Globus Toolkit to do high-performance computing clusters," says Charles Fitzgerald, a Microsoft general manager for business strategy.


Global grid computing may be the "Holy Grail of computer science," but inconsistencies in its fundamental software have made it harder for more companies to adopt the technology.

The Globus Toolkit founders are addressing many business concerns, including fusion with Web-services standards.

Other challenges remain. Among them: vendor support and price.

Globus runs with limited capabilities on Windows, "a shortcoming that needs to be addressed," says Globus' Foster. The lack of a Windows version means some software vendors are less willing to rewrite code to run on a grid. Univa has to develop a version of the Globus Toolkit for Windows, says William Fellows, an analyst at the 451 Group. "If they don't do it, they're going to fail."

There are some clues to Microsoft's intentions. Microsoft Research has funded University of Virginia's Humphrey to develop WSRF.Net, which lets users share data and machines in a .Net environment. The next release of Windows may include technology to make it easier to build grids--some of which may overlap with what's coming in Globus. "Microsoft is absolutely very interested in grid," Humphrey says. But it's "not sure where GT4 is going and not sure about WSRF."

Despite such reservations, Foster says the investments his group has procured from tech's table setters should wake more IT managers to grid's potential. "There's a reason why they're spending that money," he says. Grid computing may not be the computer industry's Holy Grail yet, but soon business-technology execs may not need to make a leap of faith to embrace it.

Continue to the sidebar:
Cost Factor: Licensing Concerns