Chinese Balloon Shot Down Galaxy S23 Ultra: Hands-On Netflix Password-Sharing Crackdown Super Bowl Ads Google's Answer to ChatGPT 'Knock at the Cabin' Review 'The Last of Us' Episode 4 Foods for Mental Health
Want CNET to notify you of price drops and the latest stories?
No, thank you

Top gurus convene to improve Linux

Programmers at the top of the Linux hierarchy gather in an effort to make the arrival of the next version of the operating system quicker and less haphazard.

The programmers at the top of the Linux hierarchy gathered last weekend in an effort to make the arrival of the next version of the operating system quicker and less haphazard.

About 30 top programmers--including Linux founder Linus Torvalds, and Linux luminaries Alan Cox, Stephen Tweedie, David Miller and Donald Becker--gathered in San Jose, Calif. to hash out the future of the next version of Linux. The Kernel Summit likely will become an annual event because there's no longer enough time for the old process--informal gatherings at trade shows--said summit organizer, VA Linux Systems employee, and fellow Linux programmer Ted T'so in a conference call Tuesday.

Predictable improvements to Linux have become more important as the software grows from its hobbyist roots into a business venture with the involvement of companies such as Red Hat, Transmeta, Caldera Systems, SuSE and IBM.

Though these companies fund Linux development to make improvements, they also depend on those improvements arriving on time. The technology market slump has removed what leeway Linux companies had.

The corporate trappings of Linux get stronger with each passing year. IBM, Advanced Micro Devices and EMC sponsored the summit, while Linux companies paid for many of their employees to attend.

Most of the improvements the programmers discussed had to do with improving Linux for use on servers, the Unix clone's stronghold. But speeding up development of the heart of Linux, called the kernel, also is a priority, T'so said.

Torvalds released the current kernel, version 2.4, three months ago but about 13 months later than he originally hoped. Torvalds had exhorted programmers to focus on creating the stable, production version 2.4 instead of adding new features into the development version 2.3, but the schedule still slipped.

"It's always more fun to work on the new stuff than stabilize the existing kernel," T'so said. "We were very ambitious in the number of things to accomplish in 2.4. Therefore, it took a lot longer than we thought it might take."

Those at the CNET's Linux Center summit didn't set a schedule for the 2.5 developer version or the stable 2.6 version, but T'so said they hoped to make it shorter than the 2.3 series. "Everyone agrees the last development cycle, which was a little over a year and a half, was too long," he said. Six to nine months, though, isn't enough time to get much done, he said.

"We're going to try to keep the 2.5 series shorter than what we did in the 2.3 development time frame, but at this point it's premature to come up with any dates," he said.

T'so said programmers are working on cleaning up the 2.4 kernel. "The 2.5 kernel series will probably start in a few months," he said. "We're still trying to iron out the last wrinkles in the 2.4 series."

One difference this time around is that new features in the 2.5 development kernel likely will be "backported" to the 2.4 kernel as they become stable enough to use, T'so said. This technique was used with the older 2.2 kernel, which benefited from 2.3 enhancements to support features such as USB (universal serial bus) connections.

New and improved
Though the focus is on 2.4, programmers at the summit naturally also discussed many features for 2.5, T'so said, most of them for servers but some for desktop machines. Among those features:

• The never-ending quest to get Linux to work faster on higher-end systems stuffed with CPUs.

Some are working on changes to the networking software that will let network devices take over some of the processing burden from CPUs. Red Hat's Tweedie discussed ways to use storage systems with more than 2 terabytes of space. And others are working on "asynchronous input/output," which speeds up storage operations by letting a computer's disk subsystem decide what sequence is best for reading and writing information fastest.

The 2.4 kernel works well on four-processor systems and even eight- or 16-processor machines with some computing tasks T'so said. "Obviously, there is more work to be done in that area," he said.

• Incorporation of new Internet communication standards into Linux.

Among those coming standards are SCTP (Stream Control Transmission Protocol) for reliably sending streams of information over the Internet; version 4 of the NFS (Network File System) originally developed by Sun Microsystems; and the ECN (Explicit Congestion Notification) protocol to deal with Internet bottlenecks.

• Incorporation of the ACPI (Advanced Configuration Power Interface) standard for power-saving measures used on laptops.

The older Advanced Power Management standard is being phased out, T'so said, and companies such as Transmeta are interested in ACPI for portable devices. "Major architectural changes will be required to do ACPI sleep and hibernation modes," T'so said.

• New and better file systems.

There are several new file systems that add to Linux "journaling" features that log transactions to make recovering from a computer crash less difficult. Of these, the ext3 system likely will show up in the 2.4 kernel as well as the 2.5 version.

"Though (ext3) is technically in beta, it's pretty solid, and we expect it to get merged into 2.4," T'so said. We held back ext3 because we really wanted it to be perfect. Stephen Tweedie and I have much higher standards" when it comes to file systems," he said.

Other journaling files systems--XFS from SGI, JFS from IBM, and ReiserFS from SuSE--will be coming later, and customers will be able to chose among them depending on the various advantages of each.

• Support for computers such as SGI's that use hundreds of CPUs.

Though there was some discussion of whether the new kernel should support the computer designs of SGI. These NUMA (non-uniform memory access) designs can accommodate more than 1,000 CPUs, but their operating system must be adjusted to accommodate the fact that sending information from a CPU to memory takes different amounts of time depending on how close the memory is to the CPU.

It's not likely that NUMA support will show up in the 2.5 kernel, though, T'so said. "There are very few people who can afford NUMA," he said, speculating that support might arrive in the 2.7 development version.