Linux is a UNIX-like operating system that is largely (and increasingly) POSIX-compliant (1). It was first written by Linus Torvalds (see The Rampantly Unofficial Linus Torvalds FAQ ), who started in 1991 with the idea of learning the capabilities of an 80386 processor for task switching. Originally named Freax , it was originally hosted using the Minix operating system.
Linus' complaints with Minix were that it was written:
As an academic/educational OS
To be portable (8088, 68000, Sparc) using only the lowest common denominator features.
Both of these design assumptions seemed crippling, sacrificing performance for academics. The assumptions for Linux have changed slightly since then; portability now is a goal, and it is certainly not simply an "academic" requirement for software. It notably moved up on the priority list when Digital Equipment Corporation (now a Compaq company, now an HP company) gave Linus an Alpha-based system; today there are commercial enterprises actively selling and supporting products based on the various ports to IA-32, PowerPC, MIPS , Alpha, and ARM architectures.
Since that time thousands of people have contributed patches, fixes, and improvements to the kernel, adding improved performance and support for many devices and file systems, as well as an increasing variety of platforms.
Linus' fame has regrettably resulted in there being something of a "personality cult" surrounding him, with kids that I'd have to term as "worshippers."
It's obvious that the "Linus personality cult" has got to go.
The "cult" began fairly early when an FTP site administrator hated the name Freax and insisted on using the more "egotistical" name Linux. Of course, the "worshippers" have gotten younger and more clueless over the years...
The historically original and still the primary platform supported is that of IA-32 (aka"Wintel-PCs") but there are now ports to numerous other CPUs. For more architectural details see Linux Architecture.
Linux is being developed under the GPL, and thus is freely redistributable in source codeform. It is particularly notable as an operating system truly developed over the Internet. Many of those who have participated in itsdevelopment are in different countries, and most have never physically met. This "peer review" at a source code level on an exceedingly wide scale has led to it becoming a remarkably featureful and stable system.
Eric Raymond has written a widely read essay on the Linux development process entitled The Cathedral and the Bazaar. He describes the way Linux kernel development uses a "Bazaar" approach where code is released as quickly and as often as possible, and that this permits extensive outside input that provides rapid improvements.
The "Bazaar" approach may be contrasted to the "Cathedral" approach used by systems like GNU Emacs "core." The "Cathedral" approach may be characterized as resulting in more "beautiful" code being released, unfortunately it is released far less often. Unfortunately, it also offers less opportunity for people outside the "core group" to contribute to the process.
Successful Bazaar projects do not necessarily involve opening the code "wide open" for everyone to play with; at the design level, the"Cathedral" approach is fairly appropriate. But once debugging the code, it is valuable to "open up the bazaar" in order to have a large number of people to find different sorts of errors. If they can contribute fixes, that is most wonderful. But if they at least locate problem areas in the code, that is still helpful to the coders.
I have written a further essay entitled Linux and Decentralized Development, which describes the way that Linux efforts are "distributed," and some of the merits and demerits thereof.
There is an experimental set of OS versions (currently the 2.5. x series) where version numbers steadily upwards on a weekly basis; the "stable" edition (e.g. 2.2.10) changes only as a result of bugs being found and fixed in the "experimental" series, and hence does not change nearly as often.
Linux users acknowledge system bugs, and quickly work to resolve them. As a result, security problems tend to be very quickly solved. There obviously is no guarantee that all users will immediately fix their systems if they are not being affected (or don't notice that they are affected) by problems, but certainly fixes are quickly available, sometimes distributed across the Internet within hours of being detected. Fixes are often available more quickly than commercial vendors such as Sun, HP, and IBM can even get around to acknowledging that there's a problem.
This is in stark contrast to the behaviour of some companies, e.g. - Bill Gates Claims "Microsoft Code has No Bugs" At least, there are none that Microsoft cares to fix. Statistically, Microsoft apparently determined that the vast majority of "bugs" represent situations where users are not using the software correctly. The remaining very few problems, caused by actual errors, are statistically insignificant as compared to the number of other kinds of problems, and thus can safely be ignored as "not real problems."
Work is ongoing on creating highly stable Linux systems; properly configured Linux kernels with suitably configured software on top of that commonly run for hundreds of days without any need to reboot. Ensuring high availability requires careful choices of hardware, discussed at the Linux-HA Project Web Site.
There are complaints that "Linux is always changing." The complaint misses the whole point of its existence. Development on the Linux kernel will stop when people are no longer interested in "building and enhancing" the Linux kernel. So long as new hardware devices such as video cards are being created and people come up with other neat improvements to make to Linux, development work will continue.
The project may eventually end as a result of there being some fundamentally better platform for "kernel hacking," or because Linux becomes so heavily patched that it becomes unmaintainable. I'm not sure that we're that point yet. Moreover, there are plans in motion to make Linux even more modular than it is now, with various schemes for moving services out of the base kernel and into "user space."
The announcement of a Debian/Hurd effort suggests an alternative; the FSF's Hurd kernel, which runs as a set of processes atop a microkernel such as MACH, may provide a more readily "hackable" system for those people that aren't quite up to fiddling with changes to the Linux kernel.
Mach provides a "message passing" abstraction that allows the OS to be constructed as a set of "components" in somewhat the style promoted by CORBA.
By running atop a microkernel, Hurd is imagined to be more readily debugged than the Linux kernel since system components may be stopped and restarted without the need to reboot the system. Mind you, with more simultaneously-moving components, one might alternatively expect more complex interactions, a greater risk of race conditions, and more complexity in capturing debugging information...