Getting with the Grid
What do you get when you cross a peer-to-peer (p-to-p) network (mostly used for trading pirated music and pornography) with a highly redundant server cluster (mostly used for mission-critical applications in Fortune 500 companies)? The answer: grid computing.
What do you get when you cross a peer-to-peer (p-to-p) network (mostly used for trading pirated music and pornography) with a highly redundant server cluster (mostly used for mission-critical applications in Fortune 500 companies)?
The answer: grid computing. Ever since Seti@Home burst onto the scene in 1997, pretty much everyone with a desktop PC at work or home has been aware of how unused computing cycles can be used for various distributed computing projects. Despite the continuing race between Intel and AMD to deliver faster, more capable CPUs, far more computing cycles are spent idling between requests than actually doing anything.
Today's local-area networks (LANs) often have interconnect speeds that rival yesterday's supercomputers. In effect, a corporate network is a supercomputer, lacking only a mechanism for coordinating the activity of all its nodes. Today it doesn't matter that Bob's computer is sitting idle while he's at lunch while Amy, in the office next door, is tapping her fingers waiting for Excel to finish crunching numbers. When her hard disk fills up with those monster spreadsheets, the 20GB of empty space on Bob's system isn't going to help at all. While the two systems are connected in the sense that they can access the same resources, they do each other no good when it comes to computational power or storage.
Grid computing aims to change this, and its backers believe that its impact on computing could equal or exceed that of the Internet. Problems that are insoluble today could be a piece of cake in a grid-enabled world, while at the same time decentralized storage could offer a whole new level of performance and reliability to corporate networks.
The grid concept is really just an adaptation of the tried-and-true cluster concept. By moving the edge of the clusters out of the server room, grid aims to deliver many benefits of clustering to user devices.
Of course, every new technology promises to change the world, wash your car and balance your checkbook. However, I think grid really does have the potential to deliver on its big promises. So do many industry heavyweights, including IBM, Cisco, Microsoft, Sun and Hewlett-Packard.
Powering Up Grid
Today, grid computing is very much in its infancy. However, its promise is so compelling that some companies are already rolling out early implementations. For instance, Pacific Life Insurance this year went with Entropia's DCGrid platform to get supercomputer-level computational power for its risk modeling system—without the expense of a supercomputer.
Grid computing isn't just for stuffed shirts and those with pocket protectors. Sony has announced that PlayStation 3 (PS3), scheduled for release in 2005, will use grid computing to deliver a huge performance improvement over PlayStation 2. While it's unclear exactly how grid is being applied in PS3, IBM's involvement in this initiative implies that it's more than just marketing hype.
If all of this sounds a bit like the p-to-p applications out there (think Gnutella), it's no coincidence. While it's clear that some central coordination is needed for grid computing, maximizing performance and redundancy means using a p-to-p model to at least some degree. The issues of scalability and security that first appeared in p-to-p applications are also cropping up for grid computing.
Finding its Bearings
Like many new and revolutionary technologies, grid is largely being held up by constant revision and improvements to the lowest layers of the technology. A few years ago, it was enough of a triumph for researchers to learn how to efficiently divide up an application's workload for distributed computing. Today, debate and development center around the standards that grid-enabled devices will use to communicate with each other. In this era of XML and Web services, most efforts are headed in the standards direction.
The Global Grid Forum (GGF) is at the center of these efforts. Membership in the GGF reads like a combination of the Fortune 500 and the venture capital section of Red Herring magazine: On one hand there's Boeing, Raytheon and Ford (in addition to the aforementioned computing giants); on the other hand, you have the likes of Entropia and VA Linux Systems—relatively small but significant upstarts.
Like most technology development or evangelist organizations, the GGF operates several types of working groups, each one creating the protocols and definitions necessary to the overall grid initiative. Fortunately—and you might want to knock on wood here—it looks like grid will arrive as a unified, open standard that will be conducive to vendor interoperability, the lower costs associated with competition, and quicker fixes for security issues.
We're just beginning to contemplate the business implications of grid computing. Larger companies will likely find that they don't have a need for all of the processing power of their own grid, and could generate revenue by leasing time on their “supercomputer.” Likewise, grid may have an effect on the type of workstations that businesses buy. Keep your eyes open as grid moves closer to real-world deployment.