Grids moving beyond science

A few years ago, Daniel Kaberon never would have considered deploying any of his applications on a grid-computing environment. Now he can’t imagine life without the distributed technology.

A few years ago, Daniel Kaberon never would have considered deploying any of his applications on a grid-computing environment. Now he can’t imagine life without the distributed technology.

Today, Hewitt Associates, a Lincolnshire, Ill., human resources consulting and outsourcing firm where Kaberon is director of computer resource management, is running an application that calculates pension benefits on a grid of Linux-based, two-processor, 2.8-GHz blades from IBM.

Kaberon says the grid, which was put into production last fall, has reduced costs associated with each calculation by more than 90% by moving the processor-intensive application off the mainframe and onto less-expensive blade servers. “And we improved the performance of the application demonstrably,” he says.

Kaberon already has plans to put another application on the grid and says he expects other applications to follow.

While grid computing remains largely the purview of scientific and research communities, Hewitt is part of a growing movement among mainstream organizations to exploit the technology to improve the efficiency of IT resources and boost application performance. Grid computing, which has received increased attention as talk heats up around various utility computing technologies, basically uses surplus computing power to run distributed or parallel application workloads.

Even today, corporations using grid technology tend to be manufacturing companies; pharmaceutical companies doing computationally intensive modeling and simulation work; and financial firms focused on processor-heavy data analysis applications, experts say.

A survey of 180 companies last summer by research firm Summit Strategies found that 4% of respondents had implemented a grid, and 12% were currently evaluating the technology. Half of the respondents said they didn’t have a timeline for when they might deploy grid technology, and 18% said they expected serious grid evaluations to be at least a year away. About a quarter of the respondents said they expected grid to be either extremely important or very important in their IT infrastructure during the next three years.

Vendors such as IBM, HP, Oracle and Sun all view grid computing as a key part, if not the foundation of utility computing, where corporate resources are pooled as a unit that grows and shrinks in response to business demands. But when it comes to business applications, grid computing is still in its early stages.

“You’ve got this basic challenge that not every flavor of application and information processing benefits from the CPU load balancing kinds of characteristics of grid computing,” says Mary Johnston Turner, vice president and practice director at Summit. “If you’ve got particularly transactional applications, there’s really no business case right now to put them into a grid environment.”

But the trend is moving in that direction. The Global Grid Forum (GGF) has worked to create specifications to make it easier for transaction-focused enterprise applications to operate within a grid. The Open Grid Services Architecture, developed by the GGF, is now focused on integrating grid standards with Web services.

Grid-focused companies made inroads last week in San Francisco at GlobusWorld, a conference organized by The Globus Alliance, which created the open source Globus tool kit designed for building grids. IBM, along with Akamai Technologies, The Global Alliance, HP, SAP, Sonic Software and Tibco, introduced three Web service specifications designed to support event-notification infrastructure for Web services that can be integrated with grid computing.

The new specifications provide “a foundation for the Open Grid Services Architecture,” IBM said in a statement.

Industry observers say adoption of standard specifications such as those IBM and its partners put forward will be the catalyst for more widespread adoption of grid computing. The ability to manage workloads and monitor and charge back for resource usage, as well as enhanced security capabilities, will be important for corporate deployments and for specifications to continue to evolve, analysts say.

At the same time, grid vendors are rolling out consulting and professional services to help corporate customers understand and design appropriate grid environments. Vendors also are grid-enabling other key offerings: IBM added grid capabilities to its WebSphere application server last year, and Oracle is rolling out its grid-enabled 10g line of infrastructure products, including the Oracle Database 10g, the Oracle Application Server 10g and Enterprise Manager 10g Grid Control.

But more than just technical barriers to grid computing remain. A study of 50 companies conducted last year by grid software vendor Platform Computing found that 89% of respondents considered organizational politics the biggest roadblock to grid deployments.

“Inside these corporations you have a huge IT infrastructure that’s been built up by department, by lines of business, so they become very siloed,” says Ian Baird, chief business architect at Platform Computing. “We spend a lot of time with our clients silo-busting because you have people that are effectively ‘server huggers.’ They don’t want to let go and share their resources because they feel they’ll lose control, lose budget and lose priority over their work.”

But Baird says grid middleware from companies such as DataSynapse and Platform provides users the ability to manage workloads across the shared resources.

“When customers start to understand grid, understand how it works and understand things like the intelligent rules-based engines in Platform’s technology, they see they can manage and have that control, but still share resources,” he says.

Nevertheless, most businesses are taking first steps with grid. Hewitt started off with a dedicated grid running one application, instead of tapping into unused CPU cycles on servers or desktops running other applications.

“There was all this stuff about service-level management that I was a little concerned about that says, if I’m using this whole grid made up of scavenging cycles from other computers, how do I know at any given time how much horsepower is going to be available for me and wouldn’t that cause an erratic performance environment,” Kaberon says.

Last year, Hewitt began working with IBM’s Design Center for e-Business on Demand to create a grid architecture Kaberon felt comfortable with. Now he says he’s ready to consider putting a composed print application on the grid and see how sharing the resources will work.

“The issues we need to work through with [grid middleware vendor] DataSynapse are what’s the right way to manage the work so this much horsepower is always available for the calculation engine, this much for the print application and then let’s keep a pool of resources so that whoever needs them can come and get them,” he says.