UpFront

diff -u: What's New in Kernel Development

Zack Brown

Issue #148, August 2006

Work continues on the git revision control system. Recently, Linus Torvalds eliminated the dependency on an external diff program, resulting in execution times a sixth of what they had been under Linux and a fiftieth of what they had been under Cygwin. At the same time, other folks are adding colorization support to git's diff output. In other git news, Petr Baudis, the Cogito maintainer, has been working on rename support. One of git's innovations is that renames are detected at read time, when someone wants to track the history of a file, rather than at write time, when the file actually changes. But, implementing this read-time detection is a challenge. Petr and a variety of others, including Linus, have taken it up with great fervor recently.

Several bits of kernel infrastructure are on the chopping block, notably DevFS. The option to use DevFS has been disabled since 2.6.13, three releases ago, and Greg Kroah-Hartman has posted patches to remove the DevFS code from each subsequent kernel release. The blkmtd driver, allowing memory devices to appear as block devices, is on the fast track out of the kernel. The block2mtd driver does the same thing and hasn't seen a bug report in more than a year, and in any case, blkmtd conflicts with H. Peter Anvin's klibc project, which is vying for kernel inclusion. Andrew Morton was more than happy to push the blkmtd removal patch over to Linus, without requiring any fermentation time in the -mm tree.

As mentioned above, H. Peter Anvin is pushing for klibc inclusion into the kernel. Klibc is a small, in-kernel libc, that allows certain kernel projects to exist in user space, safe in the assumption that the interfaces they need will be available when they need them. In this way, Linux continues to become more like a microkernel over time, without making the kind of speed sacrifices that have marginalized microkernels over the years. Linus expressed his desire to see the kernel gradually become more modular in this way several years ago, and rather than a massive sudden effort to accomplish it, there seems to be a gentle ongoing assumption that if code can be moved out of the kernel, this is a good thing. Apparently, klibc is one of the good-thing enablers.

Documentation is a rare and precious stone in the Open Source world. Recently, Chuck Ebbert put together some additions for the ptrace(2) man page that he'd gathered together from the linux-kernel mailing list, the source code and his own experimentation. Various folks, including Daniel Jacobowitz who had authored much of the functionality Chuck had documented, were happy to see this work, and Daniel offered a bunch of suggestions for improvement.

The 2.4 kernel tree continues to rest in “deep maintenance mode” as Will Tarreau puts it. Herbert Rosmanith had asked if TPM would be back-ported from 2.6, but it looks like that almost certainly will not happen. Willy has been maintaining a group of 2.4 patches gathered from various places, not necessarily in the hopes of seeing Marcelo Tosatti accept them into the official 2.4 tree, but more to enable 2.4 users to keep up to date on various drivers, without having to jump all the way into a 2.6-based system. With 2.4 apparently stationary at last, this puts more pressure on the 2.6 developers to find a way to bring stability to the kernel. Recently, Linus Torvalds began insisting that new 2.6 code would be accepted only for two weeks after an official release. After that, only bug fixes would be taken. Some developers chafe at this restriction, but there does not seem to be any huge outcry against it. However, although this means that most 2.6 kernel releases will tend to be fairly stable from an up-time perspective, it does nothing to stabilize the behaviors and interfaces between kernel releases. For that kind of stability, someone will need to have a new idea.

Ingo Molnar and others have implemented “lightweight user-space priority inheritance” support for futexes, which they say represents a significant milestone toward providing true real-time support for user-space applications. The issue is very controversial, partly because—as Ingo said in his announcement—an alternative set of priority inheritance code had been “circling Linux for years” that had bad overhead problems, buggy implementations and was just a mess. Real-time support in Linux is more widely controversial as well, because of the complexity it tends to add to the kernel.

LJ Index, August 2006

Doc Searls

Issue #148, August 2006

1. Percentage of Global 2000 companies that will have formal open-source management strategies by 2010: 90

2. Percentage of companies that have deployed or are considering deploying open-source applications: 81

3. Percentage of companies that say open source has helped lower IT costs: 67

4. Estimated dollar savings at Backcountry.com, thanks to open-source order management: 150,000

5. Percentage cost of open-source Zimbra e-mail to Backcountry.com vs. proprietary alternatives: 10

6. Thousands of pages in Backcountry.com's internal wiki “knowledge management system”: 6.2

7. Thousands of unique visitors per week to Backcountry.com, as of April 2006: 100

8. Thousands of SKUs (stock-keeping units) at Backcountry.com: 80

9. Number of new daily SKUs at Backcountry's Steepandcheap.com subsidiary: 1

10. Top daily dollar sales so far (as of April 2006) at Steepandcheap.com: 26,000

11. Millions of dollars taken by Backcountry.com in 2004 sales: 27

12. Millions of dollars taken by Backcountry.com in 2005 sales: 52

13. Number of persons employed by Backcountry.com: 260

14. Percentage of Backcountry.com employees doing in-house open-source programming: 10

15. Millions paid by Red Hat for JBoss: 350

16. Number of mentoring organizations approved for Google's Summer of Code: 102

17. Billions of dollars Intel plans to spend marketing inexpensive computers in emerging markets: 1

18. Sites surveyed by Netcraft in May 2006: 81,565,877

19. Millions of new hostnames on the Net by May 2006: 7.2

20. Apache Web server market-share percentage in May 2006: 64.76

1: Gartner, Inc. (via CIO Insight)

2–14: CIO Insight

15: CMPnetAsia

16: LinuxDevices

17: Knight Ridder Newspapers

18–20: Netcraft.com

Linuxfest Northwest 2006

Steve Hastings

Issue #148, August 2006

Linuxfest Northwest, the largest users group conference in the Pacific Northwest, is held annually in Bellingham, Washington, 20 miles south of the Canadian border, less than two hours from Seattle by car. Linuxfest is put on each year by the Bellingham Linux Users Group in a joint effort with other users groups from the area. Linuxfest is always free of charge and open to all; this year, an estimated 800 people attended.

Each year, I am impressed by the top-notch presentations Linuxfest manages to get. More than 40 presentations covered topics from general interest to advanced systems administration. Presenters included people from IBM, Novell, Sun Microsystems and Oracle. Also, for the third year, Chuck Wolber held the Alpha Geek competition. There were only four time slots for all these presentations! They easily could expand Linuxfest to a two-day festival, without even needing to get more presenters.

Danny O'Brien of the Electronic Frontier Foundation (EFF) gave a presentation called “Incoming! What's on the EFF's Radar”. He briefly explained the EFF and then covered the topics that the EFF currently is most worried about. He is a very engaging speaker.

Danny O'Brien from the EFF; his T-shirt says, ++ungood

Todd Trichler of Oracle explained “Oracle Contributions to Linux”—why Oracle is good for Linux and Linux is good for Oracle. Oracle has made numerous contributions to Linux, including the Oracle Cluster File System (OCFS2), which is now in the main kernel. Oracle has its own team of Linux kernel developers, and this lets Oracle provide complete support—no matter what breaks in a Linux Oracle system, they can fix it, period. They cannot offer that level of support for any proprietary operating system.

Ted Haeger of Novell showed off the results of “Desktop Innovation on Linux at Novell”—changes inspired by usability testing, new desktop search features (Beagle), a music player (Banshee), a photo manager (F-Spot) and a whole bunch of slick 3-D-accelerated eye-candy effects.

Ted Haeger (Reverend Ted) from Novell.

Linuxfest is hosted each year by the Bellingham Technical College (BTC), which is a perfect venue for a technical conference. The BTC culinary arts students served up a barbecued salmon lunch again this year; the weather turned rainy, but people braved the elements to get the salmon.

The exhibits room included Google recruiters; Linux-related businesses, such as Pogo Linux; users groups; Internet hosting services; free software projects, such as Ubuntu Linux and MySQL; and a computer hardware swap meet.

The day finished with the annual fund-raising raffle. Several thousand dollars' worth of donated prizes included server hosting services for up to a year.

Linuxfest Northwest 2007 is already being planned for Saturday, April 28, 2007. Be there if you can!

Picasa on Linux

Doc Searls

Issue #148, August 2006

In May 2006, Linux Journal was the first publication to learn that Google planned to introduce a Linux version of Picasa, the company's digital photo management and sharing software. This is the first time Google has chosen Linux as the first additional platform for a formerly Windows-only product. In the past, Google has expanded first on Macintosh. Here's hoping that Google Earth and other originally Windows-only Google products will migrate to Linux as well. Google runs its massive search engine on Linux servers. Although Google won't disclose how many servers they run, the company is widely regarded as the largest deployer of Linux in the world.

Picasa began as the product of a Pasadena, California company by the same name. Picasa was founded in October 2001. Google bought the company in May 2004.

They Said It

Open source is a great way for a small, hungry company to build on a sophisticated platform and get a competitive edge—an edge that's precisely tailored to their specific needs—far more cheaply than it could with proprietary software.

If nobody is dismissing you as hype, you are not loud enough.

—Bruce Sterling (at SXSW 2006)