-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
From: Jim Gray
Sent: Mon, 5/11/1998 2:30 PM
To: Nathan Myhrvold; Bill Gales; Eric Rudder; Jim Gray; Gordon Bell; Rick
Rashid; Chuck Thacker; Roger Needham; Paul Maritz; Jim Allchin (Exchange);
Gregory Faust; Dan Rosen; Greg Maffei; Charles Simonyi; Mike Murray
Subject: RE: Free software economics: The DollarOS
Nathan:
A stimulating and well-reasoned argument for priced-software.
Two things trouble me about the situation you describe:
(1) Linix is a cult that captures the best-and-brightest kids.
This is anecdotal, but I see it regularly in 10-year-olds, high schoolers,
and University faculty comments on inbound students.
The Linix cult views Solaris as bad and Windows as evil or stupid.
SUN has put the Solaris source on public view (5005 gets you the CD, and you
can apply that towards the purchase of SUN gear.) The contract is draconian (SUN
owns the rights to derivative works) but no one seems to notice that.
All of this is simply bad for us.
We have dramatically related the rules for access to NT Source, but we are
still a long way from our competitors.
Solaris is âscaleâ player in the OS space.
Linix is a huge training ground and experimental laboratory for Solaris.
Suggestion: We need to find an analog to create a âcultâ of core windows
developers.
(2) I assume Windows is heading for a on dollar OS: the DollarOS.
Currently we sell about 30 million W95, NTs, WINCE per year. I assume that
W98 and NT will converge
Following Mooreâs law, we will be selling 100 M WinNTs/year in 5 years. I
also hope that WINCE takes over the PDA space and so it will sell 100 M/year in
5 years. These do not seem radical numbers to me. But, I also expect that each
disk drive and NIC will want to run either NT or WinCE. (right now Wind River is
the OS of choice here and it is quite pdcey and not very good). All these
âperipheralsâ will have controllers that am supercomputers and will have 128MB
of DRAM in that time frame. So, all those microprocessors are going to want an
OS, a network stack, security, management, and TOOLS. This trend could drive our
volumes up 2x more to about 400 M units/year in 2003 (a wild and optimistic
guess). But the trend requires CheapOS (say one dollar for a disk controller
that costs 30$ to make).
I assume WinCE is our move towards CheapOS and LiteOS but I also guess that
we will face a WinCE â NT convergence in that timeframe. All these numbers are
HUGE volumes. If the volumes go up 100x then prices could rationally drop 100x
(which is about right for the dollar OS rather than the 25$ OS). So, I think we
might be a VERY high volume and low-cost OS company in five years.
Jim
Jim Gray, Microsoft Research, 301 Howard St #830, SF CA 94105
tel. 415-778-8222 fax -8210 Gray@Microsoft. corn
http://research.microsoft.com/barc/qray (Intranet. http://msr/groups/barc)
â Original Messageâ
Plaintiffâs Exhibit
Comes V. Microsoft
MS-CC-MDL 000000032123
HIGHLY CONFIDENTIAL
From: Nathan Myhrvold
Sent: Sunday, May 10, 1998 9:55 PM
To: Bill Gates (billg), Eric Rudder; Jim Gray (gray), Gordon Bell (gbell),
Rick Rashid (rashid); Chuck Thacker; Roger Needham; âpaulmaâ, Jim Allchin
(jimall), Gregory Faust; âDan Rosenâ, Greg Maffe, âCharles Simonyi (charless)â;
âMike Murrayâ
Subject: Free software economics
Free software, or even software distributed in source form is a current
cause celebre. Netscape, Linux and others are doing it. Is this some sort ot
trend?
There are two answers â one immediate and pragmatic and the other
philosophical.
The pragmatic answer is that much of the âtrendâ towards free software is
very likely due to the novelty of the Internet. In the early days of the PC
industry there was a period when âSharewareâ. People like Jim Button and others
developed word processors, communications programs and all sorts of other
software on a shareware basis. It was distributed by BBS systems (which were
themselves powered by shareware). Richard Stallman started the epic saga of Gnu.
This early stage of free software had its adherents â and it too was a hot topic
in the trade rags. Back then you could write a pretty decent word processor with
one or just a couple people so it could be supported on a shareware basis. Over
time this became less and less true, and shareware diminished in importance to
just a few areas. In any area of consequence a for-profit, paid for software
companies using normal channels took over the bulk of the market.
Fast forwarding to the present, the appearance of the Internet caused yet
another temporary situation where a small number of people could create a very
competitive product. Mosaic, Apache and Linux florished during this period, just
as shareware did in the earlier era. Once again you could have a software
product based on a very simple task â like serving up a file in response to
HTTP.
I believe that most of the growth in Linux can be traced directly to these
new Internet application areas where commercial software companies have not yet
created products so demonstrably superior that they have obviated the niche
occupied by free software. Although people claim that Linux is growing, my bet
is that if you subtract out web servers and related new niches, the growth is
much more modest.
If this is so, then we can expect that over the next several years
commercial software companies will displace them because web servers will become
far more interesting and complex. Straight HTTP will get ever more complicated
and extended. This has happened with a vengence for the browser, and it will
occur on the server too. If nobody can beat Linux and Apache with commercial
products, then shame on all of us in the industry!
So, the pragmatic answer to the free so,are trend is to say WHAT TREND?
Netscapeâs gambit to distribute source is too recent to say whether it will work
or not. In the case of some other products there is a definite growth trend, but
this is explained most simply as a transient effect where the Internet has made
simple software viable, and along with it free software has become viable. This
explains Apache, Linux and many other free soltware packages. If history is a
guide, over time somebody will find a way to make money from these categories
and the âtrendâ will reverse. However, even if it doesnât there is no evidence
to date to suggest that this is something fundamental about software economics.
It is more about the historical artifact that HTTP and other protocols are so
simplistic.
As much as I Iike a smug dismissal like this, it does raise the
philosophical question. What do we know about the software economics for free or
cheap software? I will concentrate on system software, but much of the same
arguments apply to applications.
Today, system software is priced between 5% and 10% of the price of the
hardware it runs
MS-CC-MDL 000000032124
HIGHLY CONFIDENTIAL
on. On smaller volume platforms that take a proprietary systems software
approach (Apple, Sun, SGI and so on), the true percentage is higher, given that
they must amoritize some fixed costs over a smaller base. This is compensated by
the fact that proprietary hardware margins often subsidize the software.
That level supports the world we know today â which has (my guess) a few
tens of thousands of people worldwide writing core operating system software.
Several times as many write system-ish software which has a price level that is
linked to the core operating system. It also supports the current user base of
150 to 200 million users.
Consider two alternative worlds â one in which the operating system is much
cheaper â say 0.5% to 1% of hardware cost, and another in which it is tens times
what it is today- at 50% to 100% of the hardware cost.
If you made this switch instantly, there would be some shocks to the system,
but instead letâs look at the steady state condition â as if the pricing models
above had been long standing traditions.
In CheapOS world, many fewer people would be working full time on system
software, because there would be no revenue to support them. Features and
functionality which support the current user base would consquently be lacking.
Which means that the user base would be much smaller.
Thus the total number of people working on system software is nonlinearly
smaller â revenue wise there would be 10X fewer systems programmers per PC.
There also would be many fewer PCs. Which means fewer variety of peripherals and
other aspects of the industry. The number of systems developers would be reduced
from our present world by much more than a factor of ten.
CheapOS world is a place which has a tremendously smaller user base,and a
tremendously smaller computer industry. These days the tech sector drives the
economy, but that wouldnât be the case in CheapOS world.
Linux fans and other supporters of âfreeâ software might have some arguments
against this.
First, they might say that there would be millions of developers dorking
with the free source. The problem is that those incremental improvements done by
small scale developers would not be available to the market as a whole, because
there would be nobody to integrate and test the results. As we know, integration
and testing does not scale gracefully. You can let a million people hack your
code, but gathering the improvements together so that each customer can get the
benefit of ALL the work is a mammoth task. We have a ratio of developers to
testers of 1:1, so for millions of developers youâd better find millions of
testers.
But itâs actually worse than that, because the integration and testing needs
to coordinate MUCH more than the developers do. You might be able to live with a
million distributed developers (albeit with a lot of wasted effort), but
integrating all their work is a single task, because any line of code may
conflict in some unexpected way with some other line of code. This means you
must have a highly coordinated effort of a million integrators and testers. Iâve
made it sound extreme by saying millions, but the same logic works for other
numbers â you need highly coordinated effort for integration and testing.
Presumably this is one of the things behind the Netscape free source code
move (assuming they think this way). At any rate it is the thinking behind my
suggestion a couple years ago at an exec retreat to do something similar with
our browser. The slogan was âlet a thousand browsers bloomâ- in my notion by
letting people develop extensions on top of the browser rather than hack the
source.
MS-CC-MDL 000000032125
HIGHLY CONFIDENTIAL
Hereâs how it works. A thousand browsers spring forth from a thousand
developers. However, amidst this field of low flowers would be only one tall
tree â the scale player who could afford to do the integration and testing. In
my version, where the customization would occur mainly above a set of binary
components, the value created by a scale player in those components could flow
to many of the customizers. Netscapeâs source based model is even more harsh â
those who hack the source will have no way to integrate new Netscape releases
automatically. In this model the free source lets the developers pick up some
fringe or niche markets which Netscape wouldnât get around to addressing.
However, it comes a a very high cost because changes to the Netscape code base
will cause the developers to re-integrate their work. Meanwhile Netscape itself
remains preeminent because they are the scale player within the context of their
source. If that isnât enough, I suspect that the fine print on the terms of the
source license also puts the fix in more directly.
Even if you could get enough coordination to integrate and test all that
stuff, there is another problem that an economist would point out â all that
work doesnât happen for free. The volunteer army of Linux developers and the
hypothetical integration and testing center, have some value on their time.
Calling it âfreeâ software is bogus â instead of paying money to a software
vendor, there is a hidden cost in the time of the users, or their organizations.
If you account for the total cost (including all the small developers, or the
cost of the users making their own mods, or the cost of users finding bugs
rather than testing finding them) then âfreeâ software can get pretty costly.
Not only do you have to account for the costs â you must also recall that
efficiency in software development depends on scale. Many years ago I wrote a
memo about leverage in the software business. If you write software for
yourself, then for every dollar you spend, you get a dollarâs worth of software
(assuming you are competent). If you have a company with N customers, it can
afford to spend a lot more on development than the cost of of the software to
any one user. So, from the userâs perspective you get a lot of leverage. For a
typical Microsoft operating system the ratio is over a million to one â for $100
you get software that cost $100 million (or more!) to develop. This miracle in
leverage means that you have software much better than you could possibly afford
to develop for yourself. It occurs, of course, because of the low marginal cost
of producing intellectual property.
If you take the same number of developers and spread them across many small
software houses or end user development, you start losing the leverage because
each individual dollar spent on development gets distributed to only a small
part of the market. The million to one leverage is the flip side of the massive
integration and testing effort to create a uniform product. You ONLY get lhe low
marginal cost of distribution to incremental users if all the users get the SAME
product. Which means the features for those users must be integrated and tested
together.
There are other scale effects as well â management is an example. An
organization can concentrate very skilled talent â if every user hacks their own
OS, very few of them will manage the work as well as a professional would.
Netscape has every economic reason to attract better managers than any of the
companies hacking on their browser source.
There are various other defenses one can mount for CheapOS world. Many of
todayâs systems programmers compete with each other because making an operating
system is a good business. In CheapOS world you could postulate some sort of
socialist ideal where the OS is some Linux like public domain thing. Thus even
though CheapOS world has many fewer developers, they are all behind one product.
There are many problems with this. Competition is intense in todayâs
operating systems and that does drive a lot of innovation â losing that
competition is not necessarily an efficiency improvement unless something else
starts to motivate the developers â for example they
MS-CC-MDL 000000032126
HIGHLY CONFIDENTIAL
could compete in enhancing the public domain OS, or in supplying new
features as middleware on top of it. Here again we run into the integration
problem â competiting enhancements in effect split the market, even if based on
common source (the history of UNIX is a classic example). Competing middleware
does too.
Maybe a set of super smart programmers would write operating systems for the
good of humanity. Some undoubtedly would (Richard Stallman is an example), but
its hard to have confidence in this occurring at the necessary scale The Soviet
Union ran the experiment for 70 years and had a rather unambiguous result. I was
just in Russia, and I can attest to this.
So, CheapOS world is a nightmare. The system software industry is sucked
dry, undermining the foundations of computing. It surely is not an idyilic
vision of the future. Since users vote on such actions with their purchases, I
think that it is unlikely that we will see it. Only some draconian force â such
a misguided government â would put the industry in this sad situation.
What about PriceyOS world where the OS costs 10X as much as it does today?
In the short term it would be a windfall for software companies, but the
short run is not the relevant topic. The current discussion is limited to the
steady state after PriceyOS world had been around a long time. I believe that
you could create a stable situation where we could spend 10X more money
developing our products than we do today, in return for 1OX greater revenues
(and the same margins). We would need to do this because our various OS
competitors would have access to those kind of revenues too, and over time we
would reach some equilibrium where we all spent like crazy developing software.
It may seem wildly unrealistic to say that we could scale up development
even if revenues supported it. How could we possibly manage teams 10X as big?
Easy is the answer- just look at history. In thje last decade we HAVE increased
our development spending by more than this amount, even if you look at a single
product. Each step happened incrementally, and we adjusted to it. If somebody
had told us then that weâd scale up by 1OX we would never have believed it â but
we did it all the same.
If we had grown in up in PriceyOS world, over time we would have done the
same thing. Indeed if you reject the notion of being able to spend 1OX more on
development, lhen you must think that the industry will die soon, because in a
few years we will be spending 1OX more than today (exponential growth is like
that).
We wouldnât be the only ones to adapt; various parts of society would have
to change too. Increasing development resources by 1OX would put talent at a
premium. Which means salaries would rise, creating more incentive for CS grads.
The industry would reach out to international developers in India and elsewhere
even more than today. Again, this will happen in the near future â as China,
Russia, India and other large countries get more software work, and thus more
incentive, it is only a matter of time before there are 1OX more programmers on
earth than today.
Even in the US there would be increases. It might seem crazy to thinkthat
there could be 1OX as many programmers, until you look at the statistics of how
damn many lawyers there are. It might take a while to reach a new equilibrium,
but somethnig like PriceyOS world is almost certainly possible.
This amazing increase in development resources would get spent lots of ways.
Lots more features. Lots better integration, testing and lots fewer bugs Lots
more niche support for features that do not make the pdodty cut. We might have
teams of human code optimizers tweaking assembly language. We might have whole
concurrent teams turning releases sooner because they would develop in parallel,
There would be many ways we could spend
MS-CC-MDL 000000032127
HIGHLY CONFIDENTIAL
a vastly larger development budget.
Actually, we have an existing example of this. Intel has much higher profit
margins than we do on its CPUs, and takes a much larger chunk of the total cost
of a PC than operating systems do. Given their amazingly high revenue per PC, in
a real sense Intet is living in PriceyCPU world â a close cousin to PriceyOS
world. And, as a matter of fact, Intel does ALL of the things mentioned above.
They do support for niche features (like MMX and other examples). They do a lot
more hand optimization of critical circuits than we do for critical code. They
do lots more simulation and testing. They have multiple implementation teams for
different chip families. We used to have two main thrusts with Win 9X and Win
NT, but increasingly we are moving to one, because we feel that we cannot afford
huge parallel efforts. Intel does however, and with Intel style revenues per PC
we could, and in think in the long run would, tend to support at least three
parallel efforts of comparable complexity. Just as Intel does.
You might say that this is because they are in the hardware business, not
because they have much higher revenue. Surely software is different. I think
this is backwards. Hardware can be done with low margins â as many chip makers
know all too well. In those cases you donât find Intel style spending. There are
some obvious differences between hardware and software, such as the capital
requirements for building FABs. However, I think that in most
cases even these can be handled without breaking the basic analogy that a
PriceyOS company would be a lot like Intel.
If you reversed the revenue roles, and gave Intel the same revenue per PC
and margins as Microsoft, and Microsoft the Intel revenue and margins, life
would go on. Intel would produce fewer new chips, and Mooreâs law might happen a
bit slower. In compensation, some of the development spending at Microsoft would
go to size reduction and performance enhancements.
PriceyOS world would have much better software than we have today, just as
we have much better software today than we had in the past. It seems pretty
good, but there is a major flaw. You canât keep raising the total cost of buying
a PC without eventually effecting the market. Demand elasticity is not infinite,
and a world where the OS cost 100% of the hardware cost, would in effect double
the price of a PC, which would cut the potential market and rob the whole
process of some of the scale effects and leverage that is so powerful.
So, it would seem that the optimal pricing for system software is some
balance between demand elasticity with respect to price on the part of users,
which strives to keep the cost low, versus the benefits of greater development
spending. The optimal value is thus somewhere between where we are today (where
the system software cost is demonstrably not high enough to impede user demand)
and something a bit more expensive. In this view, PriceyOS world is suboptimal
because a 1000% increase in the price of the OS would put a damper on demand.
Or would it? We have been innundated in the last couple years with studies
of the Total Cost of Ownership that claim that the actual cost to most users of
PCs is vastly higher than the cost of the hardware and software combined by a
large factor- say 2100% of the basic hardware cost (i.e. $14K a year for three
years for a PC that costs $2K).
You could increase the cost of the OS by 10X and it might not matter if the
consequence is that TCO was lowered a sufficient amount. Even a 10% reduction in
TCO would let you fund a 10X increase in lhe price of software and still give
the end user a substantial discount versus today. It entirely reasonable to
postulate that if you increased the software development budget by 10 TIMES that
we could reduce TCO by 10%. Indeed, the NT5 plan of record should reduce TCO by
more than 10% for a much more modest increase in development costs. Yet we
arenât planing anything like a 10X price increase for NT.
MS-CC-MDL 000000032128
HIGHLY CONFIDENTIAL
There are many flaws with the TCO studies, but nobody disagrees that owning
and using a computer does have hidden costs. So even if the numbers are too
high, the principle holds that the cost of system software (or other software
for that matter) is a very small percentage of TCO. As a consequence, small
improvements in TCO could allow you to price software much higher â and still
give customers a BETTER deal than the get today, even if the numbers are not
what Gartner claims.
Above I argued that distributing source code to customers for them to hack
for themselves was inefficient, because the âfreeâ software just incurred lots
more end user costs. These costs were large in aggregate, and furthermore are
inefficient because distributed development lacks scale leverage and canât
concentrate talent.
Just like TCO. The work that end users, system integrators and other perform
as part of their TCO chores is just as distributed and just as ineffecient as if
they were all developing the system. In fact, if you count TCO, then our CURRENT
software pricing is in fact a CheapOS scenario!
From that perspective, we would be much better off charging a lot more for
software, IF (and only if) in return it could solve all the TCO hassles. Those
hassles would be better managed, and software which solves them could achieve
vast scale leverage, rather than being the domain of the PC guy down the hall.
So, if anything recent experience suggests that the trend should be toward
more expensive software, not cheap software!
Of course, there are several objections to this. The first and biggest is
lhat users MIGHT percieve TCO differently than retail cost. Suppose that there
were two PCs in the local shop â one costs $2K and has todayâs level of hassles,
the other costs $15K (with a lot going to software annuity) over three years,
but has radically lower TCO. Would people buy it? They certainly should if they
figure TCO costs, but it does not mean they would. Humans are notorious about
subject valuation â which is why nuclear powerplants are much more scary to
people than driving to work â even though the commute is demonstrably more
dangerous.
Even if in the long run it would be more efficient, the bootstrap period in
which the perceptual view would dominate would be hard. Big MIS organizations
that track TCO might be the first to switch, it might take a very long time for
home users or other parts of the market to catch up. One reason is that TCO
average do not apply to many customers. It is much better for them to buy a
flaky machine if their pattern of use does not expose the flakyness, or does not
do so often enough that it is a problem.
More generally, there is an interesting phenomenon that ANY system software
pricing model tends to get stuck in a rut. If we lived in a competitive CheapOS
world, it would be hard to raise prices to the current level. Similarly, it
would be hard for us to raise prices today, even if you could prove that the
benefit (from TCO or other means) was worth it to society Over a long period of
time it could be done, but only with a great deal of effort. The world tends to
equilibrate around a certain level of software pricing, and resist changes even
if they are for the better with respect to the economics of BOTH customers and
vendors. Change is not impossible, but it has to be done in a very clever way.
Billâs famous open letter to hobbiests posed this dilemma â saying that heâd
love to have people pay for software because then he could afford to hire some
more programmers and make the product better. Itâs time to admit that it worked
out pretty well!
However. today you could equally well write the open letter saying that we
wished people would be willing to spend a lot more money on software so we could
cure the TCO problems,
MS-CC-MDL 000000032129
HIGHLY CONFIDENTIAL
or promote better ease of use, or be able to take on new tasks and advanced
features and functionality. Even as far as we have come to date, the main theme
of the letter is still valid software has yet to bottom out in its potential. In
the current climate I donât advise posting such a letter, but the amazing truth
of the matter is a world that invest substantially more in software would
probably be a lot better for everybody.
Nathan
MS-CC-MDL 000000032130
HIGHLY CONFIDENTIAL
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
iEYEARECAAYFAkqvOpoACgkQU4xAY3RXLo7PUwCfaDtNR7ji/JBQAXUbxoHEgJKa
IZEAn2DO11CJ04mRurYlCGs4x3d6NlI2
=FZ89
-----END PGP SIGNATURE-----
|
|
