August 11, 1999 12:00 AM PDT
Linux founder outlines vision of software's core elements
- Related Stories
Oracle to unveil Linux-based software lineAugust 6, 1999
Beowulf computing method makes business inroadsAugust 5, 1999
Linux improvements should appeal to desktop marketJuly 31, 1999
The future of Linux therefore holds power management and several other features aimed at making the "open source" operating system better for ordinary computer users, Torvalds said today at a keynote address at the LinuxWorld Conference and Expo today.
Future initiatives planned for the next version of Linux include improvements intended to support a range of technologies, including sound cards, universal serial bus (USB), PCMCIA cards, and plug-and-play features. Work to incorporate DVD drive support is being hampered by trade secrets in decrypting DVD information.
"It's not just the high end. We're trying to make Linux look good" on notebooks and low-end systems as well, Torvalds said.
Glimpses into the future of Linux aren't just important for Linux hobbyists. With computer companies such as IBM and Red Hat staking their future on the operating system, Torvalds's prognostications are the closest thing to a Linux product road map.
The founder of Linux is the single most powerful person in the software?s realm, but Torvalds doesn't control everything that becomes part of Linux. For example, he doesn't concern himself directly with the Linux user interface, nor even with the core elements of the software (called the kernel), delegating a lot of the work to trusted people.
But while he may not be omnipotent, his words have the ring of gospel to many in the Linux development community.
Torvalds?s vision is critical to its development plans. So far, the increasing presence of Linux in the product lines of computing companies and their reliance on scheduling hasn't caused problems, he said. "I've been very happy with how the commercial people have worked together with the technical people," Torvalds said.
Torvalds recently reworked the Linux development strategy for the next release of the current kernel of Linux. Where the version 2.2 took two and a half years to develop, the next version, 2.4, should take less than a year and should arrive before 2000.
Torvalds earlier had planned for the next version to be 3.0, due in a year and a half or so, but employing that more ambitious and longer-term strategy frustrated developers and made upgrading difficult. "I've been happier" with the faster, smaller model, he said.
"It may work so well that we may do a 2.6 and then a 3.0," he said.
But as a result of the faster and more modest philosophy, several fancier, high-end features won't be appearing until later upgrades of the Linux kernel.
Among the features that won't make it into the next version are a journaling file system--a feature that keeps track of changes in the operating system to make recovering from crashes faster and easier. Windows NT and most commercial versions of Unix have that feature, and Torvalds said at least two development efforts are under way to bring it to Linux.
Another feature that won't arrive is the support for Intel's upcoming 64-bit ?Merced? chip, which won't be ready in time for 2.4. "The Merced port is apparently going really well," he said, though he's not among the people who have signed Intel's nondisclosure agreement to gain access to the inner workings of the chip.
Intel said today that the first version of Linux for Merced, under development at a multi-company project called Trillian, is expected to be unveiled in the first quarter of 2000, along with a few publicly available Merced prototype computers the open source community can use to prepare Linux for the general availability of Merced later in 2000.
Clustering technology, which makes it easier to gang together multiple computers, also will have to wait another year or two, Torvalds said.
The biggest change at the high end will be in better support for systems that use multiple processors. With the introduction of the current kernel, Linux now "scales" to computers using four or eight CPUs, but much of that work was laying the groundwork for other improvements that will arrive in kernel 2.4.
"I'm sure our friends in Redmond [that is, Microsoft] will work really hard and act as our beta site and find some problems, but they'll have to work a lot harder at it," he said.
SGI is planning to endow Linux with its methods of using dozens of processors, a technology called non-uniform memory access, or NUMA. However, Torvalds said he doesn't foresee that effort as becoming part of mainstream Linux any time soon.
"Almost certainly, a kernel that runs on 64 CPUs is going to be fairly different from the one that runs on two," he said. It's possible such features could arrive with the 3.0 version of the kernel, he added.
Just because it won't be part of the official release under Torvalds's control doesn't mean that NUMA for Linux won't arrive, he said. "Different vendors are trying to specialize in different markets. SGI seems to want to work that...upwards. I think that's a good strategy," he said.
Another feature destined to arrive after 2.4 is support for PCI devices that can be swapped out without having to shut the computer down. That feature is significant for computer makers wanting to assure that their computers will stay up and running as long as possible.
"I don't think we'll have ?hot-swap? PCI for 2.4, but at least we'll have the infrastructure there so people can start working at it," Torvalds said.
With regard to managing power better, Torvalds said he's particularly interested in a system that is able to respond very quickly, shutting off and restoring power to the CPU hundreds or thousands of times each second so that even the briefest of idle moments don't waste power unnecessarily.
Support for USB is tricky, he said, because different devices rely on different standards to communicate. With 2.4, support for keybords, mice, cameras, and some scanners likely will arrive. "The more esoteric devices probably will not be supported," he said.