UpFront

diff -u: What's New in Kernel Development

Zack Brown

Issue #153, January 2007

Michael Halcrow has submitted some patches to add support for public key cryptography in eCryptFS. Overall, folks like Andrew Morton seem to be in favor of this; although Andrew points out that there already is key management support in the kernel, and that perhaps the existing code should be extended to support eCryptFS's public key features, instead of creating that support afresh. But, Michael feels he's going in the right direction, and in spite of how any particular implementation details will resolve themselves, it does seem as though public key support in eCryptFS has arrived.

Alon Bar-Lev has extended the kernel boot-command-line length from 255 characters to 2,048 characters to accommodate all the stuff that's been piling into the command line in recent years, such as module arguments, initramfs, suspend and resume, and more. Unfortunately, it's become clear that one cannot simply extend the kernel command line. The command-line code is written in assembler and has such poor design and odd code behaviors within it that simple changes turn out to require bigger fixes. But Andi Kleen, H. Peter Anvin, Alon and others have taken this as an opportunity to clean up the whole mess. So, that's exactly what they're doing. It may delay the migration from the 255 to 2,048 character boot command lines, but it probably will open up other doors that have not yet been considered.

Writing user-space PCI drivers has been an insane process, according to Greg Kroah-Hartman and Thomas Gleixner. So they decided to do something about it. Thomas wrote up some infrastructure code to rein in the whole process, and Greg added his own touches. Now they've released the code, and a bunch of folks, including Andrew Morton, have begun piling on to get it into shape for actual inclusion into the kernel. The code already seems poised to become a generic user-space driver subsystem, not only for PCI drivers. So naturally, a bunch of people are considering possible names for the subsystem—everything from User Space Driver (USD) subsystem to Framework for Userspace Drivers (FUD) subsystem. Personally, I'd like to see a subsystem called FUD. Meanwhile, folks like Manu Abraham already are chomping at the bit to see this thing implemented fully, as it would have made some work he did with Andrew de Quincey go much more smoothly.

Neil Brown has been frustrated by the sheer number of ways it is possible to feed configuration parameters into the kernel in recent years. Between sysctl, SysFS, module parameters, kernel parameters and (in a hushed whisper) ProcFS, he doesn't know which thing to use anymore to configure some random module he's writing. He has asked for help and guidance. The discussion that followed may not have led to a definitive answer for Neil, beyond Horst von Brand's recommendation of sysctl, but it did manage to get Oleg Verych to talk about his new configuration interface, called etab (short for External Text and Binary). The etab interface stores configuration in key/value pairs, and according to Oleg, may be useful in many parts of the kernel where configuration is done.

Joerg Roedel has implemented the protocol defined in RFC 3378 to allow Ethernet packets to be tunneled through IPV4. As Philip Craig pointed out, iproute2 already exists and would be a logical place to add Joerg's features. Joerg has agreed with this, but says he did the implementation separately to gain experience. Once the code begins to stabilize, his plan is to add it to iproute2.

Intel's Arjan van de Ven has announced the first release of the Linux-ready Firmware Developer Kit. This open-source Intel initiative involves a set of tests to see how well a system's BIOS will interact with Linux. Hopefully, says Arjan, this will help BIOS developers ensure that their systems continue to interoperate with Linux. Intel also is hopeful that developers will hop on board and start feeding bug fixes and support for additional BIOSes to the upstream sources.

Google Offers Code Search—Are Koders and Krugle Feeling Lucky?

Doc Searls

Issue #153, January 2007

In journalism, we say, “three examples makes a trend”. In business school, professors teach that three competitors make a market category. Both tropes now apply to code search, since Google has jumped into a market pioneered by Koders.com three years ago and expanded by Krugle.com in the middle of 2005.

I asked Chris DiBona, Google's top open-source guru, about differentiation. He replied, “We are more comprehensive by an order of magnitude, and I think we give a faster, smarter experience. Our dupe-detection is really cool. You can almost instantly see which routine is more popular/used in the world (search for btree or some other common algorithm).”

Koders and Krugle are hardly standing still, of course. And, they can now press their advantages around the edges of a large market presence. For Koders, those include algorithms optimized for code searching and results ranking, search filters, an API so other services can access the search index, and an Enterprise Edition that searches behind company firewalls. For Krugle, those include iterative searching, search of related non-code documentation, ties of metadata to code, and a notes function for comments on (and linkage to) code.

Those, of course, are subsets of current offerings by all three services, which are sure to evolve and change even more as competition heats up and programmers become more involved.

LJ Index, January 2007

Doc Searls

Issue #153, January 2007

1. Number of top ten most reliable hosting providers in September 2006 that run Linux: 8

2. Number of top ten most reliable hosting providers in September 2006 that run “unknown”: 2

3. Number of top ten most reliable hosting providers in September 2006 that run Windows: 0

4. Number of top 50 most reliable hosting providers that run Linux: 23

5. Number of top 50 most reliable hosting providers that run FreeBSD: 6

6. Number of top 50 most reliable hosting providers that run Windows: 12

7. Number of top 50 most reliable hosting providers that run “unknown”: 5

8. Number of top 50 most reliable hosting providers that run Solaris: 4

9. Number of sites surveyed by Netcraft: 97,932,447

10. Results in a search for “linux” at Google codesearch: 4,280,000

11. Number of lines of code indexed by Koders.com: 424,227,372

12. Results in a search for “linux” at Koders.com: 179,222

13. Results in a search for “linux” at Krugle.com: 700,529

14. Billions of dollars in sales for the Linux server submarket in Q2 2006: 1.5

15. Percentage increase in Linux server submarket sales for Q2 2006: 6.1

16. Linux server shipment percentage growth for Q2 2006: 9.7

17. Blade server sales percentage increase for Q2 2006: 37.1

18. Blade server shipment percentage increase for Q2 2006: 29.7

19. IBM's percentage share of blade server sales for Q2 2006: 39.5

20. HP's percentage share of blade server sales for Q2 2006: 38.9

1–9: Netcraft.com, October 8, 2006

10: codesearch.google.com

11, 12: Koders.com

13: Krugle.com

14–20: International Data Corp.

A Little Linux

Doc Searls

Issue #153, January 2007

It's been a year since Mirus announced Koobox, a new line of desktop PCs that come loaded with Linux. The first offerings were standard tower configurations, starting at $299 US, pre-loaded with Linspire's latest distro. Then, in summer 2006, the company added a Mac-Mini-like unit with a mouse/keyboard/speaker bundle for $399.99 US (after a mail-in rebate). Since then, Mirus has been adding other Mini models, scaled upwards with faster CPUs, bigger drives and features like DVD+RW.

Mini Koobox

Mirus is a subsidiary of Equus, a Microsoft Platinum OEM and Gold Certified Partner, yet calls itself “The Largest Whitebox System Builder to the Channel” and was named by CRN as number 1 out of the 50 system builders. It'll be interesting to see how it does.

They Said It

I'd much rather pay for DRM-free music than get copy-protected music for free.

All the creativity, customer whims, long tails, and money are at the network's edge. That's where chipmakers find the volumes that feed their Moore's law margins. That's where you can find elastically ascending revenues and relentlessly declining costs.

The supermodel couldn't find a rat to eat.

—Said by somebody at the Freedom to Connect conference