UpFront

diff -u: What's New in Kernel Development

Zack Brown

Issue #129, January 2005

Markus Lidel has been named the official Intelligent Input/Output (I2O) maintainer. Designed by the I2O Special Interest Group, I2O is a hardware specification that allows hardware to offload I/O processing from the CPU, raising the performance of I/O on that hardware, while at the same time reducing its impact on the running system. Markus is a relative newcomer to Linux kernel development, first appearing in the kernel changelogs in March 2004 with some minor I2O patches for the 2.6.1 kernel. Alan Cox, as with so many other projects, was the de facto maintainer at the time and accepted many patches from Markus over the following few months. By July 2004 and the 2.6.8 release, however, Markus in turn was accepting I2O patches from other developers. In August 2004, Warren Togami nominated Markus to be the official maintainer, and after an acknowledgement from Andrew Morton, Markus updated the Maintainers file in September 2004 to list himself as the official I2O maintainer. Shortly thereafter he initiated a rewrite/reorganization of the I2O code, which was accepted into the 2.6.9-rc2 kernel, in accordance with Andrew's new looser policies on large changes within a stable kernel series.

Kernel development is often fraught with controversy and dispute. Recently the pwc driver, supporting Philips Web cameras, erupted in a dispute between the driver maintainer, Nemosoft Unv, and Linux Journal columnist and kernel lieutenant Greg Kroah-Hartman. To support various hardware, a hook to allow binary modules to link into the kernel had existed for a long time in the driver, against kernel policy. Eventually, Greg insisted on its removal, and Nemosoft asked Linus to remove the whole driver from the kernel. Now, under the GPL, the driver author has no legal right to insist on this, but Linus Torvalds felt it was important to honor the author's wishes, especially if the code in question was about to be unmaintained. This turned out to be an unpopular decision with people like Alan Cox, who saw the licensing issue as much more cut-and-dried. Nemosoft had released the code and couldn't just take it back. To illustrate his point, Alan said he might as well ask Linus to remove all of Alan's contributions over the years, which would in fact remove quite a significant percentage of the kernel and completely devastate the entire project. Alan's point simply was that it made no sense to honor developers' requests in such cases. After much arguing, Luc Saillard reverse engineered the binary module in question and posted a new version of the pwc driver, without the disputed hook.

David Engebretsen decided to stop being the PowerPC maintainer and has been succeeded by Paul Mackerras and Anton Blanchard, both of whom have contributed many PPC patches over the years, along with Benjamin Herrenschmidt, Tom Rini and many others. David maintained the PPC port as part of his job at IBM and led the team that did the original Linux port to the PPC64 architecture.

Jeff Garzik has created blktool, an easier, more generic version of the existing hdparm utility. Like hdparm, blktool can wreak havoc on disks if used improperly. Also like hdparm, blktool still is fairly IDE-centric, though Jeff is working on SCSI, I2O and RAID support. hdparm's IDE-centrism may stem from the fact that it was developed by Mark Lord, the original IDE subsystem maintainer back in the early days. Like hdparm, blktool provides a command-line interface to the nitty-gritty details of your disk, allowing the user to make fine adjustments in behavior, that could result in noticeable speedups. The precise interface used by blktool's command line is still in flux; in particular, Alan Cox feels that as long as a new tool is being developed, it should fix some of the problems hdparm displayed, particularly in the command-line interface. Jeff seems amenable to various alternatives, and it looks as though the final tool will have a variety of alternative mechanisms for users to get their points across. The real point seems to be that Jeff's initial attempt has found support among kernel developers, and various folks are hacking it into shape.

BZFlag: www.bzflag.org

Don Marti

Issue #129, January 2005

The name of this network 3-D tank battle game pays homage to Atari's 1980 arcade classic, Battlezone. Shoot the other team's tanks and capture their flag, or simply go “Rogue” and shoot whomever you want. There's a large player community and action happening on dozens of servers all the time.

The basics of operating your tank and shooting are easy to learn—but beware: some people are very, very good at this game.

Devil's Pie: www.burtonini.com/blog/computers/devilspie

Don Marti

Issue #129, January 2005

Isn't moving windows manually a drag? If you're running a simple window manager, such as Metacity, but you want the window matching and customizing features of a more complex window manager such as Sawfish, here's an add-on X utility for you. Write an XML config file to match windows you want to customize, and let Devil's Pie position or tweak them for you. For example, you can pin a certain application to appear on all desktops or move all terminal windows with a certain title to their own desktop.

LJ Index—January 2005


  • 1. Millions of dollars spent for Oklahoma City's new public-safety Wi-Fi system: 90

  • 2. Square miles covered by the new system: 650

  • 3. Degree of accessibility to the Net over the system by the public: 0

  • 4. Number of cities belonging to UTOPIA, the Utah Telecommunications Open Infrastructure Agency: 14

  • 5. Thousands of households in UTOPIA's footprint: 140

  • 6. Percentage of UTOPIA that's fiber optic: 100

  • 7. Reported percentage of Linux “savings potentials” vs. Microsoft Windows: 30

  • 8. Three-year savings on office apps, in thousands of Euros, for large-scale enterprises with 2,000 jobs: 525

  • 9. Three-year savings on servers, in thousands of Euros, for large-scale enterprises with 2,000 jobs: 57

  • 10. Three-year savings on content management, in thousands of Euros, for large-scale enterprises with 2,000 jobs: 32

  • 11. Three-year savings on databases, in thousands of Euros, for large-scale enterprises with 2,000 jobs: 21

  • 12. Number of enterprises surveyed for the above study: 50

  • 13. Peak gigaflops of the new Cray XD1 Opteron/Linux-based supercomputer in a 12-processor chassis configuration: 58

  • 14. Peak number of processors for the new Cray, with 12 chassis in one rack: 144

  • 15. Peak gigaflops for a full rack configuration: 691

  • 16. Price in millions of dollars of the new Cray: 2

  • 17. Speed in MHz of the MIPS-based Sha hu (little tiger) system on a chip for embedded Linux designs, ready to run Linux: 400

  • 18. Typical power draw of the Sha hu, in watts: 1

  • 19. Millions of students to whom China wants to bring computing: 200

  • 20. Number of computers China wants to deliver, per student: 1

  • 1–3: CLS Communications, over Business Wire

  • 3–6: UTOPIA

  • 7–11: Research & Markets

  • 12–16: InternetNews

  • 17–20: LinuxDevices

On the Web

Embedded systems are what software wants to be when it grows up. Challenges such as using less memory, booting quickly and staying up for a long time without updates are requirements. Whether you're doing an in-car Linux system or developing phone switches for work, our Web site can help.

  • If you're curious about what is involved in developing for embedded platforms and targets, check out Dr Richard Sevenich's new Web series, a hands-on introduction to embedded development. Part 1 (www.linuxjournal.com/article/7848) explains how to pick a target for your embedded design. Future articles will discuss hardware setup, getting familiar with uClinux, adding a GUI and both porting and creating new applications. Follow his progress on-line or grab your own target device and become part of the project.

  • We've been following Carbot, an in-car computer company, in its quest to reduce OS boot times, a major barrier in the widespread adoption of car-ready computers and applications. The ultimate goal was to get a five-second BIOS boot time. In the final installment of this series (www.linuxjournal.com/article/7857), author Damien Stolarz reveals whether the team achieved its goal—“throw away all the useless BIOS functionality, show video at the bootloader (that is, GRUB splash screens) and start playing audio as early as we possibly could in the boot sequence”.

  • Our resident telecom and carrier-grade Linux specialist, Ibrahim Haddad, spends much of his time at Ericsson Canada working on ways to improve Linux for carrier-grade systems. His article “Critical Server Needs and the Linux Kernel” (www.linuxjournal.com/article/7855) outlines four features the kernel needs for deployment on server nodes in mission-critical environments: “a cluster communication protocol, support for multiple-FIB, a module to verify digital signatures of binaries at run time and an efficient low-level asynchronous event mechanism”.

Ten Years Ago in LJ: January 1995

A UUCP connection is a good way to get your feet wet connecting to the outside world—it will teach you how to manage a news and mail feed.

—Russell Ochocki

We halted one of the PDPs, moved the data input lines over to our PC, and booted Linux.

—Vance Petree

Built with the CS/EE/Math student and serious developer in mind.

—Red Hat Software ad

They Said It

So when you find somebody smarter than you are, just coast along. Your management responsibilities largely become ones of saying “Sounds like a good idea—go wild”, or “That sounds good, but what about xxx?” The second version in particular is a great way either to learn something new about “xxx” or seem extra managerial by pointing out something the smarter person hadn't thought about. In either case, you win.

—Linus Torvalds on kernel management (lwn.net/Articles/105375)

When you think about it, it makes sense. Linux and open-source products are cheaper, more robust and more secure. Having Microsoft tell us that their products have lower TCO is like them telling us that the Earth is flat. Right-thinking CIOs know that Linux and open-source software result in lower costs and are not likely to be hoodwinked by verbal sleight-of-hand or spurious, vendor-manipulated TCO studies.

—Del Elson, Open Source in Australia, CXO Today (www.cxotoday.com)

When open-source code is properly analyzed, there's nothing better. But just putting the code out in public is no guarantee.

Current scholarly publishing models are not economically sustainable. Researchers and students have access to a diminishing fraction of relevant scholarship. But remedies and alternatives are being developed and tested.

—University of California Office of Scholarly Communications (osc.universityofcalifornia.edu)