Even though Mike Prince, vice-president and CIO at Burlington Coat Factory (BCF), is still working on implementing grid computing at his company he has already begun to reap the benefits of shifting computing power off older, dedicated machines to interchangeable, less expensive, Intel-based servers.
“I am astonished by the performance we get out of these small servers,” Prince said. “Using standard two- and four-processor systems and clustering them has a huge cost advantage over a proprietary SMP.”
Prince isn’t alone in his appreciation for grid computing, which links together autonomous computing resources such as PCs, servers and storage devices to create faster computing power or run computations that were impossible without big-gun computing power.
Worldwide grid spending is expected to skyrocket to $19.2 billion in 2010, up from just $715 million in 2005, according to estimates from Grid Computing: A Vertical Market Perspective 2005-2010, a new Insight Research report.
According to Insight, 2004 is the year grid computing moved out of the laboratory and into mainstream commercial computing applications.
It’s no wonder when you consider how grid — especially at large companies that need lots of computing power — can help CIOs cut spending, speed mission-critical calculations, scale computing power on an as-needed basis and make the most of computer assets that firms already have.
Some firms, for example, use a grid configuration to tap unused cycle time on desktop machines. Grid helps them satisfy computing needs with existing resources, thereby getting a far better return on their PC investments.
Others use grid when they need to upgrade from older, dedicated SMPs or mainframe systems.
At BCF, for example, Prince replaced his Sequent machines to manage peak loads during busy retail times, such as the holiday season. He considered a new 16-processor server but settled on a collection of standard, interchangeable x86 servers for a fraction of the cost.
“Conservatively, I would estimate the cost savings are easily three- or four-to-one,” Prince says.
But advantages go beyond cost. Some calculations BCF tried on older equipment always took so long (more than 20 hours) that they never completed correctly.
Today as the retailer migrates to grid, the same calculation finishes in a half hour, Prince said. Plus, additional computing power can be added when needed, without taking down the system for the upgrade.
“You have the option of scaling the grid without disrupting anything,” added Prince.
But of course, grid isn’t the answer for all CIOs. Small and medium-sized firms may not need all the horsepower a grid provides while others don’t use applications that can be or have been optimized to run on a grid.
“To gain any advantage from multiple resources, an application must have been designed and implemented to split its processing into independent threads of execution to allow those to be run in parallel,” wrote Gartner analyst Carl Claunch in his December report How to Choose When and Where to Use Grid Technology. “Unless the code can divide its operation across dozens, hundreds, thousands or more threads, it will not be able to scale its performance when given many resources.”
Further, some applications just won’t get that much faster. Firms that really need computing speed — such as those in the financial services and manufacturing sectors — stand to benefit most from grid’s increased horse power.
“Those applications that have substantial increased business value when they are completed faster or when they produce more-accurate, more-complete results are the ones that derive the most benefit from grids today,” Claunch explained. “Businesses considering grid computing should seek out those applications where a decrease in runtime or increase in model complexity yields obvious advantage.”