No "Forking" in Road Ahead

Software Magazine, Sept, 1999 by Philip J. Gill

The Linux Standard Base seeks to keep OS kernel on single road

For those of us who remember the many, largely unsuccessful efforts to create a single Unix standard, the news last spring that the leading distributors of Linux had banded together for the same purpose evoked a forbidding sense of deja vu. But the Linux Standard Base (LSB) is not just a good idea, but a necessary one (www.linuxbase.org).

LSB isn't so much about Linux's present, but its future, and what its backers hope it will become. For them, LSB is about preventing the various commercial Linux distributions from taking separate paths. Or, in the technical terms of the Linux developers, LSB is intended to stop "kernel forking," a process in which the Linux OS vendors start adding proprietary extensions to the common kernel to create a competitive advantage.

Right now Linux, an open source Unix clone, is on that single path. The operating system's popularity is growing at an astonishing rate. International Data Corp. (IDC), the Framingham, Mass., market research house, says Linux shipments grew faster in 1998 than any other server operating system, exploding 212% over shipments in 1997. In contrast, shipments of Windows NT Server from Microsoft Corp. grew 75%, modest only by comparison.

IDC also predicts that Linux's rapid growth will continue. Linux commercial shipments are projected to increase at a compound annual growth rate (CAGR) of 25% through 2003, compared with a 10% CAGR for all other client operating environments combined, and a 12% CAGR for all other server operating environments combined.

Of course, Linux is growing from a much smaller base than NT or other server operating systems, such as Novell's NetWare and the many Unix variants. But IDC's numbers are admittedly conservative -- IDC counted only shipments from 15 or so commercial Linux distributions -- from companies such as Red Hat, Caldera, TurboLinux (formerly Pacific HiTech Corp.), S.u.S.E., and VA Research Systems. IDC did not track the free copies downloaded over the Web from GNU and other open source sites. If IDC had added these "non-commercial" Linux distributions to those of the commercial distributions, then total Linux shipments would have grown much faster, perhaps by a factor of three to five or more.

Linux's success is due in large part to its price (it's free or nearly so), its multiplatform capabilities, and its suitability as a Web server platform.

Kernel Is Key

And unlike the early Unix market, where operating system vendors added many proprietary extensions to the AT&T source code, there isn't a great deal of incompatibility among Linux distributions. That's because all commercial Linux distributions today are based on the same kernel, the GNU General Public License (GPL), available from GNU (www.gnu.org).

"We all run the same kernel and about 95% of the same applications," says Lonn Johnston, vice president of North America operations for TurboLinux Inc., Brisbane, Calif. Johnston says there are only minor differences between Linux distributions, mostly in the installation procedures and in the names and locations of class libraries outside the kernel.

In an effort to distinguish themselves, the various Linux distributions include other free or open source software. For example, Caldera, Orem, Utah, includes copies of the KDE graphical user interface (GUI) and sample Linux applications such as StarOffice, according to Benoy Tamang, vice president of marketing. Other distributions offer the GNOME GUI and other applications bundled with their OS.

These differences don't interfere with application portability, and the idea behind the LSB is to keep it that way. A similar LSB initiative got under way about two years back, says Bob Young, co-founder, CEO, and chairman of Red Hat, Durham, N.C., but did not get very far. It was reborn this spring under a new leader, Dan Quinlan, of Transmeta Corp., Santa Clara, Calif.

Through Quinlan's leadership, LSB has become a credible standards effort with three deliverables. Under the initial plan, the LSB was to produce a reference implementation, including source code, that all Linux distributions would have to be based on to claim LSB compliance. Quinlan says many in the Linux community, particularly Red Hat and The Debian Project, a freeware commercial Linux distribution, objected.

Instead, the LSB will deliver a written specification, a test suite to verify compliance, and a reference implementation. Linux distributions will be able to submit their own source code for testing and verification, and will not have to license the LSB source code. As long as a distribution passes the LSB test suite, it can be labeled as LSB compliant.

Quinlan says there's a strong chance that a working, production specification of LSB, LSB 1.0, will be available by the time this column appears. At the very least, he adds, the written specification for LSB 1.0 should be complete by that time.

The group is also trying to minimize any differences between the LSB and current distributions. Quinlan says it's impossible to expect that existing Linux distributions won't have to make some changes, but he anticipates that they will be minimal.