| Where is the Linux kernel going? |
Jun. 20, 2007
MOUNTAIN VIEW, CALIF. -- Last week, at the Linux Foundation Collaboration Summit held at the Googleplex, some of Linux's top kernel developers discussed the state of the Linux kernel today, and where it might be going.
Among the kernel developers present were Andrew Morton, James Bottomley, Chris Wright, Ted T'so, and Greg Kroah-Hartman. About the only top Linux kernel developer who wasn't present was Linus Torvalds, the originator of the kernel.
In a panel discussion chaired by Jon Corbett, a Linux developer himself and editor of LWN.net, the group took on many contentious issues. After introductions, in which the quiet Morton unexpectedly added a note of levity by remarking that "If you don't know who I am you shouldn't be here," Corbett started the panel off by asking, "Is the quality of the current kernel (Linux 2.6.21) horrific?"
Many in Linux development circles felt that too many "not ready for prime time" features were added in the 2.6.21 kernel. Such features included the introduction of the "tickless" kernel and a new IDE (integrated drive electronics) sub-system, along with major changes in how Linux deals with ACPI (advanced configuration and power interface).
Morton fielded this one, noting, "Back in the 2.4 days, we worked on technology for others -- a technology that could be turned into a product -- and for ourselves. With Linux 2.6, we are concentrating more on creating that technology instead of creating a product. Our major feedback now comes from the vendors. They're the customers. I'm not sure I believe them, but they say it's OK."
Still, "We need to work on regressions more," Morton allowed, conceding that "2.6.21 was perhaps a bit more buggy that it should be."
Much of the work on the soon-to-arrive 2.6.22 kernel is being devoted to stabilizing these new features. Bottomley observed that the Linux kernel developers must deal with the "tension between new features and stabilizing those features." He urged users to remember that, "If you're running Fedora Core 7, then you're helping us to stabilize and test the kernel."
Bottomley continued, "You're far more likely to find the bugs than we are. Remember, kernel developers have about half of one percent of the machines that the users do." He urged users to try bleeding edge distributions, such as openSUSE or the "testing" releases of Debian and Ubuntu, and then report bugs, since "developers need the feedback."
All the developers, as well as the audience of experienced Linux users, observed that users want the best and newest as soon as possible. Bottomley said that he was "happy that we're on a short, steady release kernel."
Kroah-Hartman agreed, and noted that Linux is evolving faster than any other software project in history. "We're adding 2,000 lines of code a day; we're modifying 2,800 lines of code. This is a rate of change that nobody else has ever seen in a project."
Kroah-Hartman, who works for Novell, also noted that users are getting impatient with enterprise kernels, such as those found in SLES (SUSE Linux Enterprise Server) and RHEL (Red Hat Enterprise Linux), because they change only once every six months. He said he thinks that enterprise distributors "may switch from locking on a single kernel for a long time."
Interestingly, Wright, from Red Hat, agreed. Business "distributions will change," Wright observed.
Wright feels that all Linux distributions need better predictions of what features will be included in the Linux kernel. Ubuntu Founder Mark Shuttleworth suggested that it would be very helpful if the development community "can work out the bugs well in advance of real releases that distributors will use."
T'so noted that for ISVs (independent software vendors) and Linux distributors, the current trend in enterprise Linux of 18 month or two year development cycles is "too painful." As is it, corporate developers "have a good idea of when the major distributions will be taking a snapshot, and push features into those versions," he said. However, they "can sort of track two major distros' release schedules, but once you get beyond that it gets kind of tough."
The group also took on power management. Morton said that "power management is no longer on or off." The problem, as always, the developers agreed, is that "Linux needs specifications for devices."
Jesse Barnes, a Linux developer from Intel, added that while Intel has been putting resources into power management, "We don't have enough, and we need other vendors to step up."
Bottomley said, "We're getting everyone to look at power management strategies." He feels that while power management "will never be perfect, at least we will have the instrumentation and a lot of knobs to twiddle."
Much of the concern for improved power management comes not so much from an interest in green computing as from vendors wanting to use Linux in their mobile devices. Kroah-Hartman said, "Mobile is asking for power management. I think the servers want it too, but they don't know it."
To help this process along, Wright urged mobile device vendors to give Linux developers "more visibility into your development process."
T'so added that, "A lot of the low-hanging fruit has been plucked on the kernel side" when it comes to power management. For example, many power management problems can be solved by improving Linux ACPI support, because most of the power control is currently hidden away in devices' proprietary firmware, where Linux developers can't get to it.
The group also discussed ISV and IHV (independent hardware vendor) support in Linux. These issues were already covered in the earlier Linux-Watch article, Linux movers and shakers seek common ground.
-- Steven J. Vaughan-Nichols
Do you have comments on this story?
Talkback here NOTE: Please post your comments regarding our articles using the above link. Be sure to use this article's title as the "Subject" in your posts. Before you create a new thread, please check to see if a discussion thread is already running on the article you plan to comment on. Thanks!
(Click here for further information)
|
|
|