TWO SERVERS ARE BETTER THAN ONE!
As Tim Ouellette reported for ComputerWorld -- just like people, computers often work better together than alone.
That's why more businesses than ever are thinking about clustering computers together to ensure that applications stay up and running at all times or that large applications can run faster.
Clustering is a way to link computers together to make programs more reliable or make them run faster. There are two methods of clustering computers. The most common is high-availability clustering, which links a second computer to a primary computer to act as a backup. If the primary system fails, the second computer picks up the duties of the primary system without a pause so users don't know there was ever a problem. The second method, called performance clustering, links computers together to team up and finish problems quicker than if one computer tried to tackle the same problem on its own.
Clustering is not just for companies with lots of computers and giant data centers. It gives older servers new life by allowing them to be used as backup systems to save money. Clustering is also good for retail stores that may want a cluster on-site to make sure their cash registers keep operating during the busy holiday season, says Jonathan Eunice, president of Illuminata Inc., a consultancy in Nashua, N.H. It can also help avoid the shutdowns caused by normal computer maintenance and upgrades.
TWO REASONS TO CLUSTER
There are two very different types of clustering technologies: high availability and performance. High-availability clusters tie a backup computer to the primary system to take over the workload if the other fails.
Most of these computers can get around 99% availability on their own, analysts say, but users need better. "This sounds great until you realize that the missing 1% represents about 90 hours -- over three and a half days of downtime per year," says a report by The Standish Group International Inc. in Dennis, Mass.
The other form of clustering, called performance clustering or parallel computing, ties computers together to work at the same time on a problem, not as backups to one another.
"The movie Titanic couldn't have been done without clusters of computers doing the animation," Eunice says.
That's because technical computing needs huge applications and files that can slow down most computers. But performance clusters let each computer, or node, take a small piece of the work and get it done quicker than if one computer did it alone (see diagram at left).
But putting together a high-availability cluster isn't a matter of wiring two computers together. You must buy special software that can do the job of looking out for failures, keeping the data constant between the two computers and telling one computer to take over if another one fails. And that still doesn't guarantee perfect high availability.
A lot of clustering technology is "overblown, overhyped and under-[specified]," says James Johnson, president of The Standish Group.
He recommends caution in many cases because users could possibly have more failures from faulty clustering configurations than from typical failure problems.
Still, clustered server deployment will grow more than 160% during the next two years for high availability and scalability, The Standish Group predicts.