Articles / Is Linux Going to Reunite t...

Is Linux Going to Reunite the UNIX Market?

Soothsayers of Linux doom raise the specter of fragmentation and predict that Linux will suffer the same fate that's held back the commercial UNIX flavors. arnim rupp suggests that not only will Linux hold together, but may also be the means to reunite the UNIX factions.

Looking at the 20 years of fragmented UNIX, I was just wondering if Linux and Open Source might be the driving force to reunite the UNIX market.

Linux already gets shipped and supported by nearly all the big UNIX vendors: IBM, Sun, SGI, HP, DEC, and SCO. They all say it's just for the low-end market and reserve the high-end for AIX, Solaris, IRIX, HP-UX, Tru64, and UnixWare/OpenServer.

If you look at the list, except for SCO they're not just in the operating system market but also sell hardware. So why should every single one of them develop every single new feature of their *NIX to sell the hardware and support? Wouldn't it be more efficient to put all development into one single UNIX, one which is owned by everybody: Linux?

Let's get some rough numbers about market share from the open jobs at dice.com, searched by keywords, all data from 16.12.1999:

OS search term open jobs
OpenBSD 3
BeOS 7
NetBSD 20
OSF 61
FreeBSD 273
SCO or Openserver or Unixware 364
BSD 409
IRIX 446
"OS/2" or OS2 487
Tru64 or VMS or VAX 1214
Mac or Macintosh or Apple 2976
"HP-UX" or HPUX or "HP/UX" 3073
AIX 3122
Linux 3167
Novell or Netware 3339
Solaris or SunOS 13103
Windows NT 16428
Windows or Winnt or "Win NT" 30823
Unix 47518

Just Solaris is ahead of Linux. That's probably why Sun was the last one who felt it necessary to ship its hardware with Linux preinstalled.

Of our remaining UNIX vendors, who's the most active in supporting Linux?

SGI. Their projects include XFS, NFSv3, GLX (OpenGL for X), the 4gb-mem kernel patch, the Linux kernel debugger, raw I/O, async I/O, and lots of other cool stuff. Check out http://oss.sgi.com/.

Fewer employers look for IRIX knowledge than for dying OS/2, which probably made SGI turn their OS strategy to Linux. Their list of projects makes it look like they don't want to perpetuate a serious difference between IRIX and Linux; they're really helping big time in making Linux ready for the enterprise. Four of the features they're working on have been addressed by Microsoft as "better in NT than in Linux" [1] (XFS = journaling and >2gb files, async i/o, 4gb memory).

So what's going to be the crucial difference between Linux and IRIX which will make people want IRIX? My guess: none. SGI will someday ship and support fancy-looking MIPS boxes with Linux and their cool graphics stuff, and they'll sell even better graphics stuff than they have now, because their developers won't have to take care of some UNIX nobody wants anymore. Look at IBM, too: they ported Linux to their S/390 mainframe-line, not their own AIX.

One other reason why all the vendors should try to defragment the UNIX market: Windows NT will stop eating their market share. NT is popular because there's only one of it [2], not TruNT, HP-NT, OpenNT, NTWare, NetNT, FreeNT, NTi, NTOS, NTsServer, and NTnux. A developer who wants to support the whole UNIX market has to compile 10 times and care for 10 different compilers, libraries, directory structures, installation programs, kernel APIs, etc. The poor admin who has to take care of three different UNIXes in his company's network has to care about different program versions, different directory structures, NFS blocksizes, administration tools, etc.

A Windows developer has to compile his program exactly once and press it in winstallshield. The NT admin has to care about only one system.

Let's make a quick operating-system-fragmentation-layer-model:

No. Fragmentation layer Effect
4. Communication protocols keeps computers from talking to each other at all
3. Programming API makes it harder to port software
2. Kernel API
1. Hardware platform keeps programs binary incompatible

Fragmentation in layer 4 doesn't matter in our case, but that's what Microsoft is going to make when they win the browser war. Let's hope the mozilla beta in q1 2000 will change the situation, or that somebody buys opera and GPLs their browser.

The Programming API layer is fragmented in the UNIX world. There's mostly only ANSI C, X, and gcc, which exist everywhere, but it's still not trivial to port between *NIX.

The kernel API is the most useless fragmentation of them all. It's addressed by POSIX and UDI (http://www.project-udi.org/), but the differences between the most important UNIX kernels don't justify the problems they make. What are the big differences between the Solaris, AIX, IRIX, HP-UX, and SCO kernels?

Sure, Linux is the weakest of those kernels. It doesn't scale well over 4 CPUs, can't do >4gb ram, has no proven clustering solution, lacks grained security, etc. But it will come to the same league sooner or later, and Linux has one feature no other kernel has: the GPL, the only warranty that work which is contributed to the public will stay there and nobody can use it to only his advantage.

The reason why the UNIX market fragmented was that all the vendors wanted some features in their UNIX the others didn't have. This is useful to give customers a reason to buy a certain UNIX, but it doesn't have to be in the kernel! The difference could still be in the hardware, support, tools, applications, brand, look and feel, whatever.

Microsoft cares about Linux, but I guess they're really afraid of a reunited UNIX market. I'll go buy SGI stock now.

Footnotes:

[1]
It's interesting to note that Linux 2.2 had some performance problems when using more than one network adapter, and the three tests cited in this page used four, four, and two NICs (the Mindcraft benchmarks also used four NICs). More on the subject can be found in c't magazine.
The 37% lower TCO of NT vs. Linux is probably taken from this paper in which Solaris/Sparc vs. WinNT/x86 also comes to a 37% difference. The Sparc hardware makes up one third of the TCO.
[2]
The Windows market can also be seen as fragmented, as there's WfW, Win9X, NT, W2k, and CE. CE is not compatible at all with the rest, and Win9X apps and drivers don't always run in NT and vice versa. We'll see about W2k, both for applications and drivers.

arnim rupp has been a Linux admin for an ISP for 4 years, wrote his diploma thesis about the future of open source software (sorry, only available in German) and is soon moving to beautiful Heidelberg. Please forgive the bazillion spelling mistakes in his original writing, as he's from .de.


T-Shirts and Fame!

We're eager to find people interested in writing editorials on software-related topics. We're flexible on length, style, and topic, so long as you know what you're talking about and back up your opinions with facts. Anyone who writes an editorial gets a freshmeat t-shirt from ThinkGeek in addition to 15 minutes of fame. If you think you'd like to try your hand at it, let jeff.covey@freshmeat.net know what you'd like to write about.

RSS Recent comments

15 Jan 2000 08:08 pimpsmurf

great article.
This was truly a great article. better than any of the other ones posted. I tend to agree with the author on the topic of lets all get together and make _ONE_FREE_UNIX_. however I dont believe linux will fall because of the different distros. when you get away from linuxconf/rpm/deb/tgz files/what ever, and drop to the # sign... they are all just about alike. (except for corel/debian.) When I get on a system to configure something. I dont spend a second looking for linuxconf in the gnome menus! I login as root, and configure. With such simularity I can configure anything nessisary, quickly. so "Divide and Conquer" doesn't play a role in linux.

15 Jan 2000 10:34 maul

Agreed.
I spoke to a recruiter from TAOS last quarter who explained to me that the needs for Linux sysadmins was on the rise. I think the fact that Linux is Open Source has done a great deal to revive UNIX as a whole, because can try it out without spending money. Linux fragmentation is not as bad as some claim it is, either.

Good article.

15 Jan 2000 10:37 nacks

good article
first off good article. I have worked as an Irix administrator for several years now and find the move towards linux to be a very good move on SGI's part.

I believe that the real test for linux in the commercial market will
be when the new intel IA-64 processors are released to the market.
Most of the major unix vendors have already bought into this new
technology and I think it will be another unifying factor. It will be
much easier for vendors to support a linux distribution on IA-64 than it would be to support thier own aging unix distributions.

15 Jan 2000 15:43 jetson123

fragmentation is a myth
I don't find the UNIX market "fragmented" at all. I find it pleasently diverse: it has a rich variety of offerings that address different needs. But at the core, all those offerings share most of their APIs, tools, and administrative processes. A UNIX administrator can easily get up to speed on a new UNIX version, well-written programs that don't intrinsically require a vendor-specific capability (high performance graphics, mainframe APIs, etc.) usually port with little modifications, and programmers don't have to relearn much.

In comparison, Windows 3.x, 95, 98, NT, and 2000 are more different from one another than the different UNIX offerings. Windows CE handheld, professional and palm are even worse, having significant differences between the different platforms, and running on half a dozen different processors. In comparison to UNIX, the Windows "platform" is more highly fragmented, and it keeps changing, too.

The only thing that perhaps makes Windows versions appear a little more coherent is their use of a common processor, the Pentium. But if you stay on Pentium, the different UNIX offerings also offer a great deal of binary compatibility, even more than Windows.

"UNIX fragmentation" is a myth created by UNIX competitors that like to make the argument "the UNIX market is fragmented because it's multivendor, so just buy from us and we will solve all your problems". Microsoft, in particular, has tried this, and they have failed miserably to deliver.

The POSIX and UNIX APIs and tools are the least fragmented, widely used platform in the world, running on anything from embedded and handheld systems to supercomputers. The differences between different implementations are adaptations to particular market segments and are to be welcomed. Unlike Microsoft, who believes one size and one company fits all, in the UNIX world, there are customized, compatible, interoperable versions of UNIX for specific markets created by a diverse marketplace.

15 Jan 2000 16:31 danielvonasmuth

World domination instead of Unification

The Unix world has been fragmented for many a year. On the other
side, Microsoft has created some fragmentation of its own from
MS-DOS to W2K.

To overcome the division, attempts have been made to standardise
Unix. From Posix to XPG4 to Unix98, the standardisation process
has been slow and painful.

Linux has contributed little to promote the standardisation process and bring Unices closer together; it is Posix 1 compliant, but so are Open VMS, Open MVS, and WinNT, to name a few.
The Linux community has created its own products like the rival
KDE and GNOME to compete with 'industry standards' Motif and
Open Look. In short, Linux adds to the fragmentation, whilst getting fragmented itself, rather than heading for a union with, say, FreeBSD.

If Linux is overcoming the fragmentation of the Unix world, it is doing so not by bringing all the OSes out there closer together
through standardisation, but rather through replaciement of the many by the one, in the same way as Windows has been seen as the
single solution for the incompatibility problem, encouraging the
others to rally behind the banner of the leader.

Though diversity gives us strength, it is also our weakness.

15 Jan 2000 18:30 9opp4

Linux as a uniting force for Unix
A recent editorial made some pretty solid statements about Linux vis a vis other implementations of Unix, but(it seemed to me) to miss the obvious point. Kernels differ in no small part because of the architecture being employed, but the interface or rather, choice of interfaces is pretty consistent across Linux distributions, due not to Linus T. or his kernel, but GNU/FSF's contributions to the standard toolkit.I'm a big fan of Linux, but using GNU tools BSDs would be a similar experience to using Linux, in terms of the end user, so the real issue is the means of access to kernel services, ie shells and basic Unix tools. It might NOT be a good thing for the Linux kernel to edge out other kernels, but it WOULD be a good thing to make Unix seamless from one platform to another, whether it's BSD, HPUX, AIX, or Linux. That's a job for GNU and Unix users to aim for.

15 Jan 2000 20:49 antibasic

Linux isn't the Messiah
4gb ram, has no proven clustering solution, lacks grained security, etc. But it will come to the same league sooner or later, and Linux has one feature no other kernel has: the GPL, the only warranty that work which is contributed to the public will stay there and nobody can use it to only his advantage.>>

This one paragraph destroys Linux entirely. Don't forget that Linux is essentially just the kernel. Not all those clumsy little programs packed by each distro. It even called it the weakest of them compared to AIX, Solaris, etc. Then it singles out Linux's lack-off strong security. It even knocks on it's poor scalability and next to no clustering. And the communist GPL is a farce. Our economy is based on capitol. That's one thing the GPL prefers to ignore. A BSD style license allows for capitol venture of an open-source application but only after a massive re-write. And the line of "But it will come to the same league sooner or later" evens admits in inferiority to other unices for a second time. The only thing that Linux has going for it is it's "Everything to Everyone" mentality. Doesn't NT try to be the same thing?

15 Jan 2000 20:53 antibasic

Linux isn't the Messiah
Sure, Linux is the weakest of those kernels. It doesn't scale well over 4 CPUs, can't do more than 4gb ram, has no proven clustering solution, lacks grained security, etc. But it will come to the same league sooner or later, and Linux has one feature no other kernel has: the GPL, the only warranty that work which is contributed to the public will stay there and nobody can use it to only his advantage.

This one paragraph destroys Linux entirely. Don't forget that Linux is essentially just the kernel. Not all those clumsy little programs packed by each distro. It even called it the weakest of them compared to AIX, Solaris, etc. Then it singles out Linux's lack-off strong security. It even knocks on it's poor scalability and next to no clustering. And the communist GPL is a farce. Our economy is based on capitol. That's one thing the GPL prefers to ignore. A BSD style license allows for capitol venture of an open-source application but only after a massive re-write. And the line of "But it will come to the same league sooner or later" evens admits in inferiority to other unices for a second time. The only thing that Linux has going for it is it's "Everything to Everyone" mentality. Doesn't NT try to be the same thing?

15 Jan 2000 22:18 jruschme

The Linuxization of Open Source
The Programming API layer is fragmented in the UNIX world. There's mostly only ANSI C, X, and gcc, which exist everywhere, but it's still not trivial to port between *NIX.

The kernel API is the most useless fragmentation of them all. It's addressed by POSIX and UDI

I thought the above two quotes interesting because they remind me of my favorite gripe about Linux. To explain... back in the 80's, there were two main variants of Unix, System V (r2, usually) and BSD 4.x. The shop I worked in ran SVr2. Every few days, though, we'd see a Usenet posting of some interesting bit of software; one problem, however, usually this nifty toy would require some feature of BSD Unix (e.g., job control) and we'd be out of luck.

Flash ahead to the present. In the interim, efforts such as the Posix standards and the SVr4 "unification" have removed most of the differences. Add to that the proliferation of "standards" such as X11 and gcc and most software can be easily ported between Solaris, HP-UX, SCO, et al.

My complaint? Well, these days I choose to run a BSD variant and every few days I'll check Freshmeat and see some interesting OpenSource software. But guess what? When I actually get around to downloading the tar file and trying to build the application, I discover something interesting... some little (or not so little) Linuxism has snuck in such as a dependency on a kernel structure or on some expectation about /proc.

If we were truly unified, then I wouldn't have to run an inferior port of Wine, nor would I have to wait until someone gets around to implementing clone(), so that I can run StarOffice.

This is not unification, it's a wedge.

16 Jan 2000 23:14 dsavard

To fragment or not to frament...
Very good editorial.

However, for reasons already explained by others I do not agree on a single distribution. And even, I do not agree on a single *NIX. The
market fragmentation is always justified as long as the fragmented
market exists. And if you think Windows is not fragmented, look at
what will happen to Windows in the next two years (and I am not
talking about Microsoft anti-trust suit result). It is already planned
at Microsoft, they will fragment and specialized many Windows distributions that will be targeted at specific market segments.

A unified market is a stupid one!

Take for example the beer
market. Those that cannot make the difference between beer
and anything golden with alcohol added to it will buy anything anyway, for real beer aficionados, they will discriminate between
a ale, a lager, a pilsner, a red, blonde, stout, etc. And will pick the one that fit better their immediate needs. So, a maket that can support market fragmentation is a well educated market and a healthy one.

That's the reason I never buy Budweiser!

However, there is a market for Budweiser as well.

16 Feb 2000 14:50 lithos

Beer and Linux
The comparision between beer and linux is a interesting one they hold quite a few things in common but there are a few differences, and it's the differences that have slowed the adoption of linux.
If you take it from the perspective that linux is like beer then you can easily say that linux is not fragmented and the vast diveristy is good for linux.
If you take it from a opposing view that beer is linux software and you are linux, then you run into the problems of only being able to drink specific kinds of beer because the other 90% of brands don't work well with you. Of the remaining 10% you can only drink 1% easily because the other 9% don't have bottle caps you can easily take off.

Simply diversity is good it fosters invention and advancement. Fragmentation is bad as it limits adoption and ease of use. There is a grey middle ground that can be found, sadingly quite a few people will disagree on what that is. As it is quite a few people hate control but if you have the right controls you can both foster the good things while limiting the fragmentation. Standardization is good, one governing body for standards are good, controling everything is bad.
Examples of good standardization are things like: OpenGL, it has a standard implementation yet still evolves. Java. x86 hardware specifications, etc..
Now I do not agree with or advocate telling people that you must use xyz window manager or install software stictly to this place if it is this kind of application, or that you must include only these things in a distribution. But when someone like myself (a application developer) must struggle with a bunch of different packaging tools and different linux(s) distributions so I can get my product used it isn't fair to me, or the users.

Right now the costs incurred with the diversity of linux is just too high, there are far too few standards. Linux(s) must find a middle ground, it would be nice if they were mostly compatible with each other. A nice place to start would be with a Unified Package Manager, and before you think that package manager xyz is good just compare it to the installation/removal facility in windows. It runs under a obviously inferior OS but it's a hell of a lot better than anything I've seen on linux.

06 Apr 2000 22:53 ecsd

unification of Unix
I'm glad to see the momentum behind Linux, because Microsoft needs to be destroyed (freeing hundreds of programmers to do more useful things.)

There may be only one NT, but if it sucks, who cares?

My prediction is that by the time Linux obviously has the momentum to overtake Microsoft, BSD will be the system of choice among those in the know - say by about 2004/2005. The merger of BSDI and Walnut Creek provides a better base to try to unify the BSDs, and through that process developers should look forward to an integration of Linux and FreeBSD (Linux warm-and-fuzzy forward, FreeBSD kernel.) I'm not holding my breath for that, but I do look forward to a unification of standards.

And as always, looking forward to never having to support another Windoze machine whose OS is provided by a company that can't spell 'management', 'methodology' or 'modular'.

Screenshot

Project Spotlight

quadtree

A Thread-safe quad tree C library.

Screenshot

Project Spotlight

Xtables-addons

Additional Netfilter/iptables modules.