Grids Make Enterprise Play
Grids Make Enterprise Play
David Levines customers think playing video games on the Web with a million other people is cool. Whats not cool, however, is having your arrow miss its target during a dramatic battle sequence in the latest Braveheart game because the online gaming site youre using ran out of computing power.
Levine, CEO of Butterfly.net Inc., a Shepherdstown, W.Va., hosting company for Web-based games, is determined to make sure that doesnt happen. So hes powering his companys network with an enterprise grida network of servers that work together, seamlessly shifting processing tasks among machines as demand ebbs and flows.
The idea of linking computers to form loosely coupled grids has been around practically since systems were first networked. For the most part, it has proved practical for only technical and scientific applications, where workloads are often massive and problems can be easily broken apart. The concept of using grids for enterprise computing, however, is gaining momentum, particularly as companies continue to extend core applications outside the firewall and onto the Web, where capacity needs are hard to predict. Vendors such as IBM and Microsoft Corp. are backing the notion. Startups such as Butterfly.net are buying in. And even a few established enterprises such as aircraft component manufacturer Pratt & Whitney, of East Hartford, Conn., are beginning to wield the idea of enterprise grids with gusto, convinced they are a way to tap into vast computing resources sitting idle.
But while grids hold great potential for the enterprise, experts say most IT managers should keep a lid on their expectations for the technology. For one thing, most grid computing platforms currently lack the management software required by most enterprises. In addition, standards needed to support interoperability among grid products are not yet in place. As a result, while grids hold promise for enterprises, it will be years before they are practical for tasks that require more than just raw processing power.
"Grid computing is the next killer app or will be the catalyst for a whole new set of applications that we never seriously thought about," said Nathaniel Palmer, an analyst at The Delphi Group, in Boston. "But enterprises need to think about economies of scale before launching into it. The potential is there, but most organizations arent ready."
Getting to Grids
Getting to Grids
Hoping to make grids a reality for enterprises are a collection of leading vendors and high-powered researchers. The Globus Project, an academic collaboration that addresses the challenges presented by todays global, heterogeneous networks of computers, for example, is developing OGSA (Open Grid Services Architecture), a combination of grid computing standards and Web services standards that could make it easier for enterprises to develop applications that can run on grids.
At the same time, technology vendors are pushing the concept of enterprise grid computing, many with the expectation of using grids to offer hosted computing utilities. Although Sun Microsystems Inc., of Palo Alto, Calif., got an early jump on grids when it acquired software maker Gridware Inc., of San Jose, Calif., in 2000, other computing giants are now backing the concept as well. IBM CEO Sam Palmisano recently declared the Armonk, N.Y., company will support grid computing across the enterprise as it does Linux. IBM backed up its stated support by submitting the OGSA specification to the Global Grid Forum in conjunction with the Globus Project. Ensuring that it doesnt get left out, Microsoft, of Redmond, Wash., gave Globus $1 million this year to ensure that the organizations Globus Toolkit will run on Windows XP and the .Net Web services infrastructure.
In addition, smaller vendors that traditionally offered grid software to customers in academic and research environments are now putting together enterprise grid products. Platform Computing Inc., of Markham, Ontario, recently released the newest version of its grid architecture software, Platform LSF 5, while Entropia Inc., of San Diego, has plans to integrate OGSA into its DCGrid products. Avaki Corp., DataSynapse Inc. and Hewlett-Packard Co., among others, also offer grid solutions.
While few enterprises are ready to use grid technologies for tasks outside research and development and design, Butterfly.net is intent on proving its possible to build a commercial business model on top of grid technology. Company executives spent the past two years working with IBM to build a grid that distributes the processing of video game interactions across a network of server farms co-located at IBM data centers. Butterfly.net needed to create one grid that could handle multiple games and allocate resources to a particular game on the fly, depending on need.
The network, deployed earlier this year, enables Butterfly.net to support thousands of video gamers playing simultaneously over the Internet. The Butterfly Grid is built on proprietary software designed by Butterfly.net called Butterfly Grid MMG (Massively Multiplayer Game) Platform and the Globus Projects open-source Globus Toolkit. The Globus Toolkit allows Butterfly.net to monitor servers and distributes the processing needs of more popular games and populated areas to idle computing resources within the data center.
By running its grid software at an IBM data center, Butterfly.net can draw the computer power it needs when the number of gamers using its network spikes. The Butterfly Grid is powered by rack-mounted IBM Xseries servers running Red Hat Inc.s Red Hat Linux 7.2 and an IBM DB2 database.
This month, Butterfly.net began to combine its grid network with Web services technologies, launching Web services for programmers. Using IBMs WebSphere framework and software from CollabNet Inc., of Brisbane, Calif., Butterfly.net allows video game developers to deploy new games to the grid by automatically loading game-specific logic and art assets onto the Butterfly Grid.
Not for All Enterprises
Not for All Enterprises
While grid technologies have begun to succeed in a nonscientific setting such as Butterfly.nets gaming network, Levine and Butterfly.net Chief Technology Officer Mark Wirt agreed that the concept will not work at all enterprises. Before targeting gamers, the two considered and rejected everything from online banking transactions to concurrent engineering supported by a grid.
Grids may be too unproven to support many startup business plans, but a few established enterprises are expanding their use of the technology beyond scientific and design applications.
At Pratt & Whitney, a division of United Technologies Corp., engineers have been using grid technologies since 1992 to handle the processor workloads necessary to design rocket engines and jet propulsion systems. Using Platform Computings Platform LSF workload management software, Pratt & Whitney clusters the power of as many as 5,000 workstations simultaneously to handle high-intensity batch design computations.
Now, because grid computing concepts have worked so well in Pratt & Whitneys research and design processes, Peter Bradley, associate fellow for high-intensity computing at the aerospace company, is examining the idea of building a grid to collect and apply unused workstation processing power to mission-critical enterprise applications.
But while Bradley may be a proponent of grid technologies, he is the first to admit that the concept will be unsuitable for most enterprises today. To support a grid infrastructure, a company must be willing to take a leap of faith and invest in hardware, train IT managers and even build management tools that are missing from many commercial grid products, he said.
"There is definitely a barrier to entry with grids," Bradley said. "This is not something you can just rip the shrink-wrap off of and fire it up."
Other articles in this package: