Intelligent Enterprise

Better Insight for Business Decisions

RSS
Webcasts
Digital Library
Subscribe
Home

September 17, 2002

Distributed Decision Support

Are grids a good step - or even a likely one - for mainstream decision support?

by Seth Grimes

Grid computing provides a welcome contrast to 360-degree customer views, CRM touchpoints, and closed-loop analysis — jargon created to dress-up and differentiate mundane technologies. In a world where graphical glitz and grandiloquent marketing trump solid, reasoned design, grid computing's potential to bring unprecedented processing power to decision-support solutions — at low cost — while extending their reach, is truly exciting.

Academic researchers have been working on computing grids since the mid-1990s, joined in recent years by computing vendors that have embraced grids for a variety of reasons. First, grids offer a new scalability model for computationally intensive processing. In addition to fast processors, single-platform multiprocessing, and clustering, grids can incorporate ad hoc collections of disparate, networked machines. Just as a power grid can deliver electricity generated by coal, nuclear, hydro, and oil plants for industrial, commercial, and home use without regard for the type, generation, or use, computing grids are agnostic toward the multiple operating environments of the Internet (or even most larger organizations).

Proprietary-platform vendors, therefore, have to join the grid or risk being left out. And with basic grid protocols well established — some based on distributing computing concepts and techniques that date to the early '70s — developers are folding in Web services, which should further reduce operating-system dependencies.

Data grids have emerged more recently. The idea is similar to computing grids: create a web of Internet-accessible databases. A data grid is virtual data space bound together by standardized metadata and access methods.

Although grid computing is developing quickly, it's still years from mainstream. Nonetheless, if you're doing serious data analysis or working in an extended enterprise, the question won't be whether or not to use grids but rather "How?" "How soon?" and "How much?" The answers will depend on the nature of your computing problems and your computing environment.

Concepts and Implementations

The basic grid concept is simple: Adapt computationally intensive computing problems using software toolkits — often open source — to allow parallel processing across a network of otherwise idle machines. Although everyone bought PCs with fast chips, large disk drives, and copious memory to run bloated office applications, I can't really fault that expense because the productivity that fat PCs enable has meant a huge return on the investment. Whether alternative architectures would have done more at lower cost is academic at best, of historical interest only given the Internet-driven transformation of computing models.

You can also build a grid linking servers or other people's machines. Notably, the University of California Berkeley's SETI@home and Distributed.net's cryptography challenges both use a screen saver that fronts data-sifting processes distributed to participant machines and operating across the Internet. Both date to 1997.

Grid toolkits and products are readily available — the Grid Initiatives and Projects page (see Resources) is a great starting point for any search — as is plenty of underutilized computing power: PCs and servers sit idle much, if not most, of the time. Grid-enabled client software uses distributed computing facilities managed by grid servers. These facilities include directory, communications, scheduling, resource management, and monitoring services and security infrastructure. The net result is, in the words of software vendor Avaki Corp., "a dynamic network of computing resources that work together as a single, uniform operating environment. It can span locations and administrative domains and can flexibly support dynamically changing organizations and computing requirements."

The Global Grid Forum (GGF), an international coordinating body, hosts an open standards-development processing and a framework for cooperative development of the specifications and services by working and research groups. The GGF is shepherding significant innovations that include creating an Open Grid Services Architecture that reconciles grid computing and Web standards via use of the Web services description language (WSDL). WSDL-encoded semantics (meanings) would support "the creation, management, and application of dynamic ensembles of resources and services (and people) — virtual organizations."

Yet most analyses I've seen state that the mainstreaming of grid technology — of distributed decision support — is three to five years away. The need to restructure computing solutions is one obstacle, and corporate culture presents another significant roadblock.

On the second point, in today's climate of heightened security awareness, many organizations are loath to face the security challenges they fear are posed by distributed computing. (Meanwhile, they blithely populate user desktops with Swiss-cheese software. You know what packages I'm referring to.) Organizations nowadays also prefer buying analytic software and solutions to building their own and thus have to wait for vendor developments.

What of established software vendors? So far, software companies that provide computing platforms — operating systems and application servers — have invested most heavily in distributed, grid computing. Their approaches vary. IBM has provided distributed-computing solutions for years and strongly backs the Globus Project, an open-source, multiple platform effort. The Globus software toolkit is the product of a collaborative, team effort that operates similarly to the open-source software movement's models and now serves as a reference implementation.