Linux Clustering Extends Trend

By Lenny Liebmann

Linux delivers lots of computing power on commodity lntel processors, and it’s especially popular with Net devotees. So could Linux turn out to be the OS of choice for dot.com server clustering?

Plenty of vendors think so. Network Engines recently started to ship its XEngine Linux cluster. And, Linux systems leaders VA Research lnc. and TurboLinux lnc. are also shipping clustering solutions based on Linux.

Some prominent users are buying in: The University of New Mexico built a 128-server cluster using technology from Alta Technology Corp., which runs the Red Hat Linux OS on Pentium II processors. Nicknamed “Roadrunner,” the National Science Foundation-sponsored project gives users all over the world access to major computing power via the internet.

“We’re interested in using clusters to solve problems that would overwhelm an individual workstation,” says David Bader, an assistant professor of electrical and computer engineering who heads up the project.

“lt may be that you need more capacity, or that time is a critical factor in what you’re doing,” Bader adds. Such applications include highly complex particle physics equations and visualization of huge amounts of data generated by space observation equipment.

“With a Linux cluster, you can write an application on your own workstation and then easily transfer it over to our machine,” Bader explains. “With some other large-scale computing platform, you’d either have to have access to it for development, or you’d have to worry about porting it and modifying it after you developed it on your own machine.”

Linux clustering is not only for the rarified world of academia. Building clusters with Linux servers offers a powerful bottom line case for business users:

“The two machines I’m using in my TurboLinux cluster cost me $1,100 apiece,” boasts Michael Kuehl, president of Digital Facilities Management lnc., a Web hosting and lnternet development firm. “To achieve the same results with a commercial Unix product could cost as much as $70,000.”