Red Hat looks under Linux's hood

(continued from previous page)

software to newly created virtual machines as well as schedule jobs and balance work running across a group of servers, Stevens said.

Xen remains a relatively immature product today, however, and not one to be used by those faint of heart. By August, developers had hoped to release Xen 3.0, which is expected to include important support for multiprocessor servers. However, that version now is scheduled for release in December, according to the project Web site.

Stateless Linux
Another major push will be the stateless Linux effort to make the operating system useful for customers with large numbers of desktop computers, Stevens said. This software effort "got stalled out" for a time, but "with the new team, we're picking it back up," he said.

Stateless Linux stores the particulars of a computer user's desktop operating system on a central server and lets the person tap into it through a variety of methods, Stevens said. For example, users could have Linux installed on a PC that synchronizes personal files kept there with the same files kept on a server, so they can work offline or get faster access. Or a generic PC could temporarily be turned into a user's personalized machine by booting it using a Linux CD or a version of the operating system sent over a network.

The key to stateless Linux is creating a version of the operating system that can automatically adapt to a PC's particular hardware configuration, Stevens said.

Stateless Linux is the top priority customers are seeking from Red Hat after virtualization, Stevens said. The reason is to reduce management costs for companies with hundreds or thousands of machines, he said.

"What we're trying to do is (develop) a compelling way to drive down the cost to manage client environments," Stevens said.

SystemTap
New developer tools are on the way for Linux, too. Sun Microsystems has attracted a lot of customer attention with its DTrace tool, which finds bottlenecks by letting administrators analyze software as it runs, Stevens said. SystemTap is a newer alternative to DTrace designed to work with Linux; DTrace with Sun's own OS, Solaris.

"There was a lot of 'We like DTrace, and we want you to do that,'" Stevens said. "Now I don't hear that anymore."

There's still work to do, but Red Hat is headed in the right direction, analyst Iams said. "If you do a head-to-head comparison with DTrace, you'll see SystemTap is not anywhere near as powerful," he said. But the work so far is "an indication that Red Hat has not lost touch with the requirement of enterprise and data center users, where Linux is making a lot if its headway. A lot of the success is at the expense of Unix systems."

Red Hat also is backing development of Frysk, an advanced debugging program, the company said.

Stevens is the right person to oversee the Linux and open-source improvements, Iams said. He worked on the highly regarded clustering abilities of Tru64 Unix, developed initially at Digital Equipment Corp., before moving to a start-up called Mission-Critical Linux and then to Red Hat.

"He was one of the chief architects of TruClusters. He's got a very strong track record," Iams said. "He's one of the smartest guys I've met in my career. He understands the issues cold."

[JBorders_CNet]

Virtual goal is to run more efficiently? what?

>> All share the goal of making a server--and ultimately, a group of servers--run multiple jobs more efficiently. <<

I think Mr. Shankland needs to leave the description of what a Virtual machine does to the experts, he clearly has no business describing it himself.

A Virtual will ALWAYS be LESS efficient than running the native host operating system. The goal of a Virtual machine is NOT to run multiple jobs more efficiently (as Mr. Shankland would have you believe -- they will ALWAYS run LESS efficiently than a native host!) No, the goal of a Virutual machine is to allow flexibility, portability and reuse of machine configurations, along with the more efficient use of HARDWARE, since a single machine can now run several virtual machines, albeit in a LESS EFFICIENT manner.

I think Mr. Shankland does a disservice to those of us that understand and use Virtual machines on a daily basis. Perhaps he might quote an authority next time, rather than offering up his own, incorrect, assessment.

[Bill Dautrive]

Reading comprehension is your friiend

>> All share the goal of making a server--and ultimately, a group of servers--run multiple jobs more efficiently. <<

Read that again. And again, until you figure out that it is possible for one thing that will never match another to run more efficiently-RELATIVE TO ITSELF.

[Shankland]

Reporter responds: *overall* efficiency increases

I think we disagree on semantics, at least to an extent.

Indeed, a single application running on a virtual machine consumes more processor cycles than one running on bare hardware, although of course hardware support for virtualization in the form of Intel's VT and AMD's Pacifica will reduce that. But one of the main reasons so many companies in the computing industry are pushing virtualization, and the use of virtual machines specifically, is that you can wring more use out of a server running many different tasks.

The argument goes like this: Virtual machines increase the utilization of server hardware, and software to monitor the infrastructure will let administrators or automation software move virtual machines from one computer to another to deal with ups and downs in processing demands.

The upshot is that what took some number of servers in the past will take fewer of the same type of server in the future. Virtualization might cause an individual process to consume more computing cycles than it would otherwise, but overall, servers are being used more efficiently.