Linux 2.6 on Horizon

Linux creator Linus Torvalds claims Version 2.6's improved memory, I/O handling and scalability will be appealing to enterprise customers.

The next major release of the Linux kernel is on track for the first half of next year, with improvements in its ability to handle large amounts of memory and throughput.

Scalability enhancements, as well, will add to the appeal of the latest kernel, Version 2.6, for enterprise customers, according to Linus Torvalds, the creator and top programmer for the Linux kernel, in an e-mail exchange last week with eWeek.

While he would like the 2.6 kernel to be ready early next year, Torvalds said, "Its just too hard to predict, and it does end up depending a lot on how good the vendors are at trying to calm things down through stability fixing."

Torvalds froze the feature set last month and is planning a code freeze by years end. "Although I have a soft spot for Jan. 5 [for the freeze] ... it will be exactly 12 years since I got the PC that was to become the first Linux PC," he said.

"Ive basically cleaned up my plate of invasive patches that got in before Halloween, and things are starting to stabilize again. The biggest challenge is probably just a lot of testing at this point and checking that different drivers work in particular," Torvalds said.

There has been debate about whether the changes to the kernel are significant enough to bump the version number to 3.0, but Torvalds said 2.6 is his choice. "I dont see a huge reason for version-number inflation right now," he said.

According to Torvalds, high-end users will notice the scalability work that has been done, as well as the responsiveness to the desktop, even under heavy I/O loads. In addition, the scalability work has been "quite noticeable, so 2.6.x will bring quite a big improvement in the ability to handle huge amounts of memory gracefully and having lots of I/O active at the same time," he said.

Torvalds said he feels theres nothing he left out of 2.6, but "there are always a ton of people who have their own projects and feel differently."

One of these is Wim Coekaerts, principal member of Oracle Corp.s technical staff and its Linux specialist, who said some vital clustering abilities are still not there.

"While network failover and some I/O failover capabilities will be in the 2.6 kernel, disk volume management is something enterprise customers want," said Coekaerts, in Redwood Shores, Calif. "We would like Linux to have a Logical Volume Manager. The 2.6 kernel will have a device manager, but we need a LVM."

Paul Cornier, executive vice president of Red Hat Inc., in Raleigh, N.C., agreed.

"Making a more generic cluster file system is important to us, as is an industrial-strength Logical Volume Manager," Cornier said. "A distributed lock manager completes things. This is functionality that needs to go into the operating system but is unlikely to be found in the next [kernel] upgrade."

"I think people are grumbling mostly about the fact that they need a few new management tools, which do exist, including being backward-compatible with the old LVM setup," Torvalds said. "I suspect the user-level tools will be filled out and improved upon over the next few months."

An embedded programmer for a security technology company in British Columbia, welcomed the moves around I/O.

"The better the I/O, the better the kernel for all of us," said the programmer, who requested anonymity. "We are thrilled with what weve seen of the 2.5 development tree so far, which will find its way into the 2.6 kernel."