RealTime IT News

Forget About the Academy...Meet The Real OSCAR

When MSC.Linux, the open-source solution division of MSC.Software Corp., makes its MSC.Linux V2001 beta version available this April, it will signal a bit of a coming out party for the quiet Open Cluster Group (OCG), an organization of scientists dedicated to make clustering easier for developers.

This is because the MSC.Linux V2001 beta has been fitted with the OCG's first project, the Open Source Cluster Application Resources (OSCAR).

Simply, OCG said "everything you need to build, maintain, and use a modest sized Linux cluster is included in OSCAR." Developed for experienced cluster developers, OSCAR 1.0 was launched late February and while the organization said the OSCAR simplifies cluster building, it is fragile and requires more Linux system administration expertise -- in other words, developers to test it. And that is why MSC has spearheaded OSCAR's implementation.

OCG is led by Oak Ridge National Laboratory (ORNL), the University of Illinois' National Center for Supercomputing Applications (NCSA), IBM and Intel. MSC.Software, a founding member of the team, is the technical lead for the programming environment. Additional collaborators in the Open Cluster Group include Dell, SGI and Veridian.

Among its founding principles is the notion that while Linux is progressing, there is still a lack of well-accepted software stacks that are robust and easy to use by the general user community" to conduct mainstream supercomputing. Don't make the mistake of calling OSCAR a standard; that implies formality -- a no-no in the open-source arena. All software, of course, must be freely redistributed.

The History of the OSCAR

Tim Mattson, a senior research scientist for Intel Corp.'s Microcomputer Software Laboratory division, told InternetNews.com he would take the blame for starting the Open Cluster Group. Mattson, who had been working with clustering since the late '80s, had been attending supercomputing conferences and had the kernel of such a group in mind, but he didn't realize others shared the same view until 1999.

Upon becoming snowed-in at a clustering conference at Chicago's Argonne National Laboratory, he and other conference attendees began talking about such an open-source group dedicated to clustering.

So, armed with desire and a whole lot of expertise in supercomputing, Mattson and Stephen Scott, a scientist at Tennessee's Oak Ridge National Laboratory, set forth with the group.

"I expected it to be a real controversial meeting," Mattson explained. "Because we're all experts; it's like when you get a bunch of economists together who can't agree on anything."

Yet no such ego overload materialized, Mattson said.

"There was absolutely none of it," Mattson said. "We all just really clicked."

In April 2000, OCG made its official debut with an early blueprint of OSCAR, called the Community Cluster Development Kit, or CCDK. In a slide tutorial, OCG defined clustering as a "collection of connected, independent computers that work in unison to solve a problem."

While this seems like an overwhelmingly positive approach to highly complex problem solving via computing, the industry has been slow to adapt to the technology, as OCG said the difficulty of setting up the pieces has resulted in a lack of commercial support. Oh, that and the fact that the learning curves are steep.

But Mattson and the rest of the OCG stress that they are not trying to create a standard -- just an approach to supercomputing.

"The CCDK is a snap-shot of best-known-methods for building, programming and using clusters. It is not a new standard!" OCG states. "It will hopefully bring some degree of uniformity to clusters, foster commercial versions of the cluster software, and make clusters more broadly acceptable."

In fact, before spiraling into technical details for the cluster in this kick-off tutorial, OCG also took a couldn't-we-all-just-get-along stance by pledging its commitment to work together with others, find a common framework and define a process to reach a well-rounded solution for open clustering.

That was a year ago this April. And the Linux clustering landscape has changed quite a bit, according to one analyst.

Don't Fragment! Can't We All Just Get Along?

Dan Kusnetzky, a vice president of system software for market research firm International Data Corporation, said he was familiar with OCG's stated goals. He also suggested that its silence relative to actual businesses who are furiously preparing clustering solutions to market hasn't helped industry a bit.

"They clearly aren't getting the message out that they exist and that they're doing something important to the community," Kusnetzky told InternetNews.com. "While they quietly work away in the corner, others are running forward to claim the market opportunity. So, we're seeing a proliferation of incompatible solutions from many different companies appearing rapidly on the scene."

Kusnetzky is referring to SteelEye and TurboLinux, who have already announced products, but there are many, many more companies engaging in Linux clustering.

"Red Hat is making the Linux Virtual Cluster (LVC) software available via their high availability server product," Kusnetzky said. "SuSE is doing something quite similar. IBM is in the process of making some of the software that has made the SP such a successful product available on Linux as well. I'm waiting for Compaq to weigh in with their clustering software, Caldera/SCO to bring out Non-stop clusters for UnixWare on Linux, etc. Veritas and Legato are also working on porting their HA/clustering software to Linux."

In response, Mattson told InternetNews.com Friday that it is not OCG's contention to place in the same markets as the Linux cluster developers Kusnetzky detailed. Mattson said a fundamental difference between OSCAR and these other applications, such as Red Hat's Linux Virtual Cluster, is that OSCAR is not targeted for "high-availability" markets. Rather OCG offers their solution to select groups of scientists and developers -- not just any potential user.

Kusnetzky's list of Linux cluster creators then, raises the question then: are there, as OCG said in their kick-off notes a year ago, "too many options without clear winners?"

Mattson said there are certainly more now than there was a year ago.

"Our main goal," Mattson stressed, "is to ensure that we keep the technology from fragmenting, which is not what some of these companies are doing."