The Last Dinosaur and the Tarpits of Doom:
muq.org
How Linux Smashed Windows
cynbe@muq.org
If you are a Microsoft investor, be afraid. Be very afraid. By 2010 Windows will be as dead as
CP/M, and every Windows-based software vendor will be either supporting Linux or out of
business.
The process is in fact 80% complete: The end result is already obvious to bright CEOs, and will
shortly be obvious even to bright mainstream press columnists.
In this essay, we will skim the available evidence, extrapolate the trend, and examine some of the
mechanisms powering those trends.
How do you extrapolate an exponential?
In his 1952 essay Where to? (available in Expanded Universe) Robert A Heinlein pointed out
that there are four ways of extrapolating an exponential curve. Suppose, for example, that an
operating system "Linux" has been doubling market share every six months for a decade, and has
reached 2.5% of the desktop market. How should an analyst extrapolate the curve?
The conservative mainstream analyst will project that Linux will maintain a constant market
share thereafter, perhaps declining slowly: Linux at 2.5% desktop market share for a few
years declining to 1% within a decade, say.
The daring analyst will allow for a slow rate of growth beyond the the current level, rapidly
approaching a limit: Linux growing slowly from 2.5% of market share to say 5.0% in a
decade or two.
The very daring analyst will extrapolate a linear rate of market share growth: If Linux has
grown from 0 to 2.5% of the desktop market in the last decade, it will grow from 2.5% to
5.0% of the desktop market in the next decade, and by 2100 will have around a 25.0% of the
desktop market.
Finally, there is the mathematically correct way of extrapolating an exponential: If Linux has
been doubling every six months for a decade, the most probable extrapolation is that it will
continue to double every six months. That will put it at about 10% of the desktop by Jan
2000, about 40% of the desktop by Jan 2001, and at market saturation by Jan 2002.
Note that when Linux reached 10% of server market share, all the mainstream third-party server
software developers jumped on the bandwagon. The conservative assumption, then, is that when
Linux reaches 10% of desktop market share, the third-party desktop software developers will
likewise jump on the bandwagon. Understandably: a 10% marketshare gain is enough to interest
the shareholders. We may expect to see this avalanche effect to kick in on the desktop market
sometime in the year 2000.
Arithmetic Exercise: What is Linux's doubling time?
Datum: In 1991 Linux had one user, Linus Torvalds.
Datum: In 1998 Linux had about 16 million users. (Call it 8 or 32 if you prefer; It will
change our answer only by a few percent.)
Datum: 1024 is 10 doubling times, so a million is about 20 doubling times, and 16 million
is about 24 doubling times.
Result: 24 doubling times in 8 years is 3 doubling times a year, or an overall doubling time
of four months.
For this essay, I'll stick with the conservative figure of two Linux doublings per year: It makes
the arithmetic easier to follow. Daring minds can re-do the extrapolation with a four-month
doubling time.
Observation: The race is nearly run.
In an exponential process, each doubling is just as difficult as any other doubling. Linux has
successfully doubled 24 times so far, and needs a total of 29 to 30 doublings to achieve desktop
market dominance: accordingly, in fundamental terms, it is about 24/30ths of the way there
already, which is to say 80%.
It takes a brave or foolhardy analyst to predict that the runner will trip and fall before the finish
line when the race is already 80% over!
Beyond Numerology: Why?
It is one thing to tot up plain public facts and make elementary geometric extrapolations: It is
another to understand them. Why is Linux smashing Windows in the marketplace?
Historical Detour: The Open Standards Steamroller
For a somewhat deeper understanding, we may observe that Windows' eclipse by Linux is not an
isolated event, but in fact merely the continuation of an overwhelming industry trend in operation
for a quarter century.
In 1970, the engineering computing world was dominated at every level by proprietary
standards.
ASCII was a distant dream: Every computer had its own binary text encoding used by its own
line of terminals: Heaven help anyone foolish enough to try transporting a text file between
brands of computers. Major computer manufacturers had several incompatible text encodings:
for example IBM's EBCDIC variants and DEC with sixbit vs radix50. If you were perverse
enough to want your text to be portable across computers, you avoided risky extensions like
lower case, never mind special characters beyond comma and period.
In 1970, primitive proprietary operating systems bestrode the landscape like mighty dinosaurs:
Prime's PrimeOS, DEC's RSTS, RT-11 &tc (with VAX/VMS soon to come), IBM's innumerable
offerings, CDC's Scope and of course dominating the scientific workstation market, Apollo's
Domain.
Who would then have dared to predict the fall of such giants?
What force could topple such entrenched operating systems, backed by massive industry
investment, hacker culture and customer loyalty?
Today, of course, we all know the answer:
In 1975 Bell Labs released Unix.
Unix had no support from its creator, AT&T: Buy the magtape and don't call us. (AT&T
was legally barred from entering the operating system market.)
Unix had no support from any existing vendor: None had the slightest interest in backing,
supporting or developing an alternative to its proprietary operating systems offerings.
Unix had zero customer base: Nobody had ever heard of it, nobody was requesting it.
Unix had zero marketing: Nobody had any reason to spend money building mindshare for it.
A one-sided competition?
Decidedly: Unix wiped all workstation competition off the map in less than fifteen years.
On April 12, 1989, HP bought up Apollo at a fire-sale price, putting out of its misery the last
remaining proprietary operating system vendor in the workstation world, and the workstation
proprietary OS era was over: Unix was left alone in the workstation market.
In fifteen years, a magtape and an idea had effectively destroyed all opposition: Every
workstation vendor was either supporting Unix or out of business.
(A few of the larger dinosaurs lingered on in niche roles, of course: VAX/VMS is actually still
available if you ask for it, albeit mutated now to OpenVMS and more or less claiming to be Unix.
Possibly some retiree somewhere is still running Domain on genuine vintage Apollo hardware.
The Boston Computer Museum no doubt preserves some of these coelacanths of the computing
world.)
What was the idea that single-handedly exterminated these proprietary dinosaurs?
Open standards and consumer choice.
The same proprietary monopolies which are so very profitable for vendors are so very
expensive for their customers that, given a choice, customers pick open standards and free
competition every time.
And sooner or later some struggling vendor offers them the choice.
In the end, a mature workstation market realized that it was better served by open standards
across the entire industry, and every vendor fell into line or went out of business, all the way
from terminal vendors (ASCII or bankruptcy) to networking vendors (TCP/IP or bankruptcy) to
CPU vendors (Unix or bankruptcy).
Microcomputers: The Valley that Time Forgot
But even as St Unix was slaying the proprietary dinosaurs of the workstation world, history was
repeating itself in the Valley of Microcomputers.
In 1960 DEC introduced the PDP-1 minicomputer, and in 1963 the PDP-8, with 4K words of
memory.
In 1975 MITS introduced the ALTAIR 8800 microcomputer with 256 words of memory.
Just when it appeared that disks and high level languages were here to stay, suddenly one could
once again experience the joys of assembly coded operating systems loaded off tape!
Microcomputing has proceeded to replay the history of the workstation ever since, with a fifteen
year delay.
On this basis alone, a lazy analyst, noting 1989 as the year of the final demise of minicomputer
proprietary operating systems, could pick 2004 as a likely year for the final demise of
proprietary microcomputer operating systems -- in close agreement with the exponential trend's
prediction of Linux market saturation starting in 2002.
Another cheap prediction directly from history: Just as the death knell of minicomputer
proprietary operating systems was a lawsuit requiring that Federal minicomputer purchases
specify an open standard such as POSIX rather than a single-vendor sweetheart solution such as
VAX/VMS, the final death knell of microcomputer proprietary operating systems will be a
successful lawsuit requiring that Federal microcomputer purchases specify an open solution such
as POSIX rather than a single-vendor sweetheart solution such as Windows.
Today, the same people who picked open standards over proprietary solutions in the workstation
context are faced with the same decisions in the microcomputer context and (is anyone
surprised?) are making the same choices: Proprietary solutions in the microcomputer world are
being rapidly replaced by open standards on all fronts. Open software standards like TCP/IP are
rapidly killing off proprietary standards like IPX while simultaneously open hardware standards
like PCI are clobbering proprietary alternatives like Microchannel.
Microsoft was the biggest of the microcomputing dinosaurs pushing proprietary solutions, and
unsurprisingly it is proving the last to sink in the tarpit of open standards, just as Apollo, leading
vendor of proprietary workstation operating systems, was the last to give up the ghost.
True, like a trapped Tyrannosaurus Rex, Microsoft remains dangerous, able to rend any
competitor within reach of its jaws, even as its struggles sink it ever deeper into the tarpit.
But with the long-term self-interest of the entire customer base working against proprietary
solutions, Microsoft can only buy incremental short-term survival by selling off what remains of
its future:
To maintain the appearance of robust growth critical to its dominance, Microsoft has had to
nearly triple the real per-CPU cost of Windows over the last half-decade, from below $20
to over $50. In the short term, this is an effective way of converting monopoly power into
revenue, but the result is to increase the incentive for customers to switch vendors: Today
60-80% of IT managers say they would switch from Windows if they had a choice -- and
companies like Compaq are starting to ship boxes with Linux pre-installed. With new PC
prices dropping below $400, $50 for the OS is no longer economic: Change is in the wind.
To survive the short term, Microsoft software has had to offer steadily increasing support
for open standards: Today its applications have to support C, Java, HTML, TCP/IP, SQL
and a host of other standards beyond its control. But each time it switches from a
proprietary standard to an open one, the Last Dinosaur reduces its monopoly power, opens
the door wider to free competition, and increases the leverage customers have to demand
yet more adherence to open standards. Every such tactical victory is a strategic defeat. The
result from Microsoft's perspective is an exponentially increasing sink rate into the tarpit,
resulting in a steadily rising level of desperation sufficient (for example) to drive the Last
Dinosaur to bet-the-company confrontations with the American legal system which it would
never have considered in happier days.
To deal with competitors like AOL, Microsoft has had to go beyond merely reluctantly
supporting one standard after another, and begin actively promoting the construction of new
open standards, for example its promotion of via the IETF of an instant message protocol
standard (RVP) to compete with AOL's ICQ. Even if this succeeds in weakening AOL in the
short run, it will result in yet another industry move from proprietary to open standards,
which in the long run will move Microsoft one step closer to open competition in a free
market. No points for guessing that Microsoft wouldn't be doing this if it thought it still had
the faintest ghost of a chance of creating a proprietary standard it controlled: Microsoft is
implicitly admitting that the days of proprietary microcomputer standards are over.
To slow its sink rate somewhat, Microsoft has had to resort to increasingly heavy-handed
tactics, which may sometimes work in the short run, but which mean that is it stepping
steadily more painfully on the toes of organizations ranging from the Pentagon to General
Motors -- organizations which can and do have resources for dealing with irritants. It is no
coincidence that Microsoft misbehavior has simultaneously become of interest in venues
ranging from the US Congress to the European Union to Japan to the US Department of
Justice to any number of US state governments: Microsoft has begun buying short-term
survival at the cost of long-term hostility, a desperate bargain indeed. It is anyone's guess
what the specific result of any particular legal or legistlative action will be, but it neither
does it matter: What does matter is that vast groupings of the world's power elite now see
Microsoft as a problem to be solved, and will crank up the pressure until a solution
satisfactory to them is achieved. To them, that means being able to make commodity
purchases in a competitive market, and not being subject to the whims or self-interest of any
single CEO. The direct costs to the world economy of Microsoft's monopoly now run to the
tens of billions of dollars; The indirect costs run to the trillions. That's more than enough to
motivate mainstream action. No amount of PR posturing can hide those costs from the
people paying them, nor stave off indefinitely solutions to the problem.
Will Microsoft in the end suddenly turn to the Light Side? There is precedent in the way DEC
suddenly switched from "Unix is Snake Oil" to "We're the biggest Unix vendor in the industry".
Or will Microsoft fight with conviction to the bitter end, as did Apollo, going down with all PR
guns blazing?
We have no way of guessing, but in industry terms, it really doesn't make any difference anyhow,
except perhaps to Microsoft employees and stockholders.
Let us leave the Last Dinosaur sinking inexporably deeper into the tarpit which has now claimed
all of Microsoft's predecessors in the proprietary operating systems game.
But Why Linux?
There can be little doubt that when Windows falls, Linux will inherit its mantle: It is a simple
matter of momentum. Ask Alta Vista (say) to count hits:
Windows: 2,530,775
Linux: 502,053
Solaris: 251,513
HP/UX: 105,833
FreeBSD: 81,781
MacOS: 70,851
UnixWare: 23,386
Ultrix: 15,133
OpenBSD: 11,892
Linux is not only growing faster than any OS in history -- it already has twice the Web mindshare
of anything but Windows.
Why?
Part of the answer is of course simple chance. Had Linus Torvalds been born ten years later,
FreeBSD might well have inherited the ecological niche which Linux has claimed.
A more interesting part of the answer is the Open Source revolution: Thanks to the rise of the
Internet, microcomputing is here going on beyond a mere replay of workstation computing history
to break new ground.
Open standards make the computing world more efficient by reducing the amount of wasted
motion needed to implement a new idea: Standards cut the software environment into pieces with
defined interfaces, and only the component behind a given interface need be re-implemented.
Introducing something like SSH can now be done just by replacing a few programs like telnet,
instead of by introducing perhaps an entire new operating system.
This reduction in wasted programming effort translates directly into cheaper and more plentiful
software, which in turn produces indirect cost savings in every part of the economy leveraging
software.
Open source takes this a step further: When the source is available, instead of having to rewrite
programs like telnet from scratch, the existing source code can be merely modified, resulting in
still greater reduction in waste effort.
Proprietary ownership of source code has in the past allowed large corporations to dominate the
software market because they could modify their existing proprietary code more cheaply than
others without access to it could rewrite from scratch. This allowed extraction of monopoly rents
from customers, with the usual happy results for the seller and unhappy results for customers and
the economy as a whole.
The growth of the Internet has now put the shoe on the other foot: It has made cooperation so
cheap, and the set of cooperating Linux developers so large, that Linux now not only has a
programmer force dwarfing anything Microsoft can field (with a doubling time measured in
months), but an existing codebase dwarfing that available to all but the very largest corporations.
The Linux codebase consists of about one hundred million lines of code, which at the accepted
industry rate of $100/line, represents an effective capital investment of ten billion dollars.
For comparison, a company like Lotus or Apple has a market value of about a two billion
dollars, Oracle has a market value of about thirty billion dollars, and Microsoft has a market
value of about one hundred billion dollars.
In economic terms, if Linux were a corporation, by conventional economic metrics of book
value, it would currently be roughly the third largest software firm on the planet, behind
Microsoft and Oracle and ahead of firms like Autodesk ($1.3 billion).
In terms of sheer productive programming power, Linux already dwarfs Microsoft, and is pulling
away fast: If it were a conventional software corporation, it would have the largest effective
development budget in the software world.
Since there is no conceivable way Microsoft can match the scale of the Linux effort, there is no
way Microsoft can ever close the capability and reliability gap which Linux has already opened
up over Windows: By the time the Last Dinosaur can struggle up to where Linux already is, Linux
will be over the horizon.
Thanks to the availability of source, the entire academic and research world has become a Linux
R&D lab. For example, thanks to NASA's Beowulf project, Linux has today operational,
production supercomputers with hundreds of nodes. It is a safe prediction that within a year there
will be Linux-driven supercomputers with over a thousand CPUs -- quite possibly many
thousands.
I don't know what will come out of the research labs at NASA, MIT, Caltech and kin tomorrow
or next year, but one prediction is easy: The overwhelming majority of the advances will be built
on Linux.
New Computing World Order
The result of all this is the start of a fundamental phase transition in the way the entire software
industry works.
Innovative companies are discovering that by going with Linux, they get to leverage a software
base with a commercial equivalent worth already exceeding ten billion dollars and doubling
steadily -- instead of having to re-invent these wheels and be correspondingly late to market --
and to boot gain as an ally the largest software development team on the planet.
It's already hard to resist, and once Linux' market share passes that of Windows, it will be
irresistable and irreversable.
The future beckons!
On your way out, please leave a donation for the Dinosaur Preservation Foundation: Someone
needs to start preserving memories of the old order, before they are lost forever. |
