-----BEGIN PGP SIGNED MESSAGE-----
From: Jim Fredricksen
Sent: Friday, October 09, 1998 3:11 PM
To: Pascal Martin
Subject: FW: Much ado about Linux
thx for reminding, jim
Account Manager-OEM Sales
Tel: (425} 936-7268
Fax: (425) 936-7329 (bldg 18)
From: Raju Gulabanf
Sent: Thursday, October 08, 1998 7,49 AM
To; Chameleon Core Team
Cc: Jawad Khaki; Mike Nash; John Frederiksen; Jennifer Cioffi; Mike Oldham;
Subject: FW Much ado about Linux
Please handle with care.
From: Vinod Valloppilill (Exchange)
Sent: Wednesday, October 07, 1998 10 05 PM
To: Open Source Software/Linux (Private) (Exchange DL)
Subject: FW: Much ado about Linux
fyiâ interesting reading. direct relevance to server appliance people
From: Eric Rudder
Sent; Wednesday, October 07, 1998 9:42 PM
To: Vinod Valloppilill (Exchange), Oshoma Momoh; Oliver Sharp
Subject: FW: Much ado about Linux
just in case you guys havenât already gotten the fwdâs.
From: Nathan Myhrvold
Sent: Wednesday, October 07, 1998 9:23 PM
To: Bill Gates; Steve Ballmer (steveb), âpaulmaâ, Eric Rudder; David Stutz;
Jim Allchin (jimail), Rick Rashid (rashid)
Subject: Much ado about Linux
There has been a lot of interest in Linux as a competitor to Microsoft
operating systems recently and I thought I would add some comments of my own to
A while back I wrote a memo on free software generally in which I took a
fairly dim view of it. I wonât repeat all of it here, but the punch line is that
the only thing more expensive than commercial software with a license fee is
âfreeâ software. Having the software be âfreeâ means that you cut out only one
portion of the total cost â and as we all know from TCO studies, the license fee
for software is only a tiny fraction. Free software needs to be maintained,
tested and administered. That is where the real cost is. Eliminating the
software developer only shifts the investment around. In particular, it is an
inefficient shift because you take maintenance, testing and enhancement away
from the developer (who can build expertise, economies of scale) and distribute
it to a chaotic mix of smaller players, or even to end users. So from a basic
economic viewpoint, âfreeâ software is a fundamentally bad idea.
Only an idiot thinks that mass market software is expensive compared to any
metric â cost of hardware, cost of end user time, opportunity costâ My God, even
the cost of ELECTRICITY to run a PC for a few months to a year is higher than
Comes V. Microsoft
than the cost of the operating system! For a few hundred bucks we deliver
what â a hundred million lines of code for Windows NT? Try developing that
yourself for that price. Other commercial OS vendors do likewise. Itâs an
infinitesimal cost from any perspective.
Diving into more detail, the âcathedral in the bazaarâ theory is that the
distributed approach to software development is better. The problem with this is
that many of the key aspects of software development do not distribute well.
Testing is the simplest example â if you have a volunteer army of folks writing
their own extensions, who is going to test it all together?. Testing, as we know
from bitter experience, does not scale well. You need uniformity and control,
which you donât get in a distributed, potluck environment where everybody hack
their own features.
That covers broad theoretical aspects, but that isnât enough. Even if Linux
is on a path that ultimately bumps against economic realities, it might take
years, or even a decade for that to occur. In the meantime it could be an
important competitor, wreaking havoc with established OS providers. There are
several ways to look at Linux as a competitor.
As a desktop phenomenon, I donât think that Linux is very important. The
application set is too limited, and they are too far behind. The place where
Unix is very important (i.e. dangerous) is on the server.
This happens at an interesting time, because server based computing is
exploding. The Internet creates a vast need for new servers at every level. The
way I like to look at this is the ratio of CPU cycles (or RAM, disk whatever) on
your local machine to the CPU cycles done on your behalf on servers â i.e. R =
(Server CPU cycles) / (Client CPU cycles)
In the early days of the PC business, R = 0 because there was no servers to
speak of. These days R = 1% to 5%. This is because every PC is connected to
servers for email, HTTP or some such, each of which typically has 100 or so
users. The servers are more capable machines, but overall it still adds up to a
few percent (try doing the math on your own usage pattern).
R is going to increase steadily, approaching 1. That is a VAST change in the
The âthin clientâ, ânetwork computerâ or âcentralized computingâ fans think
that R will go to 1 by stealing functionality away from the client (i.e. doing
word processing or other core PC applications on the server). This is wrong on
several bases â instead it is new apps and expanded use of old apps (every
citizen with an email address) that is, and will continue, to drive the
penetration of new servers.
There are four significant aspects of Linux as a server OS:
1. Linux as the OS for server appliances â boxes which provide a very
limited set of Internet connectivity, email etc for a set of users in a small
2 Linux as the new Netware â i.e. a simple network OS that provides a fairly
limited set of services: file, print, SMTP, HTTP and so forth. In this mode
application availability is not a big deal because you have a fairly limited set
3. Linux as a host for Oracle or another SQL dalabase.
4. Linux as a host for large scale custom server apps â i.e. Hotmail,
Case 1 is the clearest, so I will treat this first. Simple, boring server
apps are growing like crazy because everybody needs email, newsgroups and web
pages. Small business users want something simple. I recently advised somebody
setting up an office for a couple people. A turn key, appliance like server for
basic connectivity needs would have made their lives a lot more simple.
We clearly need to compete like crazy in this space. Technically speaking
this is where a stripped down embedded version of NT would be very useful â and
as far as I know this is what we are doing.
Linux is not a particularly capable competitor in the sense that it has lots
of special technology. Ideally a server appliance OS would have a lot of
sophisticated self healing, remote admin and other features. To my knowledge,
Linux is not at that level â it is being used because it is simple and fairly
small. NT potentially has MUCH more to offer in this area. It also has some
drawbacks (size, complexity, unneeded features). If we can strip out the
unnecessary stuff, AND focus a lot of attention on the specific new technology
for this area we could be very successful.
That said we really do have to worry about this area. But we need to temper
the worry with some common sense. The whole point of a server appliance is to
open up the market to users who could not afford to use a standard NT server â
either for cost, or set up hassle etc. This expands the server market. It may
also cannibalize some degree of current
NT sales. However, it is not going to eliminate our server strategy
This is an old story. A new niche develops which is quite different from the
industry mainstream. It catches on and grows like wild fire. Because it is a new
area, the technology used to address it in the early stages is very simple, lean
stuff. So, somebody at Microsoft panics and says OH MY GOD, THIS IS THE FUTURE,
WE ARE SCREWED! They send impassioned email saying how everything we are doing
is wrong because it is not like the dead simple stuff that is being used in the
There is some Linux mail like this in the last couple weeks. Prior to that
there was some mail about how Amazon.com is the next killer app. This is healthy
and there is a lot of value in this, The mistake that is made is to view the new
area as a replacement threat, when in fact it is an incremental opportunity. A
lot of bad decisions can be made when people confuse the two. It is great to be
alert to changes in the market, but we need to recognize that most things are
secretive, not direct replacements.
Another mistake is to think that the dead simple technology used to address
the new niche is what we should be doing. This is tdcky because the people in
the new niche always talk a good game. It is easy to mistake their current state
of technology with what is desirable, especially if they are ahead of us in that
market. Although simplicity is sometimes a virtue, it is much better to focus on
what the new niche NEEDS rather than how it is currently being addressed.
Returning to the case at hand, to beat Linux in the server appliance market,
we need to add MORE technology, not less! Yes, we need to strip down NT to
reduce resources needed to run it and other reasons, but at the same time we are
stripping out irrelevant things, we had better be adding architectural support
for features that will make a difference in that market. There is no point in
competing with Linux on the basis of being a simple OS â that is fighting on
their territory and their terms. If we tried to turn NT into Linux, they would
be gaining in sophistication.
We need to address the incremental opportunity of server appliances, by
getting somebody focused VERY hard on this area â both with embedded NT but also
with thinking through what technology that market segment will need. My guess is
that in the long run server appliances are like Web TV- a product for people who
either would not have bought a PC, or who will graduate up to it eventually (in
this case, for the server rather than client).
Case 2 is, to my mind, a broader threat than the server appliance. Novell IS
the server appliance company â it started with proprietary 286 based servers for
PCs, almost exactly like the server appliances. Later they switched to software
only â but not a general OS â just software for simple network services â which
at that time was mainly file and print
Fast forward to the present day â the Internet has given a new lease on life
to simple server services. Instead of just file and print, the current
generation of simple services are driven by the Internet â HTTP, NNTP, SMTP and
various others. There is an opportunity for simple, high performance server
software that runs on PC hardware and is a bit more flexible than what you can
do with a server appliance. The market of strict server appliances will be
smaller because most users will be unable to live with all the restrictions, but
still want to use cheap PC hardware to accomplish their tasks.
If Novell had not self destructed, they should own this space, not Linux.
Alternatively, it should be Novell and Linux duking it out for this market.
Novellâs almost complete abdication to Windows NT means that the simple server
battle will be waged largely between NT and Linux.
Again, we need to gird for the threat. This is a challenge because our
strategy against Novell was that we would have the flexible general purpose
system, competing with their simple services. Just as we emerge victorious,
simple servers come into vogue again â what a bummer!
The rise of the browser was a similar situation. PCs had rich documents with
embedded graphics and images for years. All of a sudden the Internet browser was
born and was a terribly retro, backward step technologically â it was about text
with no choice of fonts, no structured graphics, only a couple image formats.
All of the complex stuff that the PC industry had done seemed irrelevant, and
the low tech stuff Netscape was using was paradoxically cool. Ultimately, the
technology treadmill of improvements was what mattered. Netscape and Microsoft
added feature after feature, and in the long run the better software development
In the long run our flexible approach to servers with Win NT will win once
again over Linux as it did over Novell. Linux will be unable to keep up with the
pace of development. Once again, I want to reiterate that the way to compete
here is NOT to try to adopt a Linux like stance â instead we need to add
technology to improve our product for this market.
Case 3 is an interesting one. Large SQL databases have taken on so many low
level functions that they are almost operating systems unto themselves. As a
result, products like Oracle are able to support dozens of operating systems.
They can easily support Linux, and are likely to do so
Our strategy in this case is again to promote product features â such as
cluster support, remote admin, ZAW and so forth to one up Linux. We have to
assume that it will be a viable platform for Oracle and others, because they do
so much of the work internally.
Case 4 is the topic of another round of hand wringing. Many web sites are
basing their custom software on UNIX, and could easily do the same for Linux.
Most current operating systems do not have much support for the kinds of
problems these folks must tackle â nor for that matter to current email systems
(designed for smaller scaleâ) or current databases (designed for fewer users,
different usage patternsâ). In the current early generation of systems like
Amazon or Hotmail their creators have to do almost everything from scratch. In
such a context, the OS does not matter very much â at least for now.
At the moment, Solaris and other commercial versions of UNIX are probably
more of a competitive threat than Linux when it comes to large web sites like
Hotmail or Amazon. In part this is due to better features, and in part because
of Sun hardware. However, we should expect to see Linux used quite a bit in
these contexts. However, looking forward my opinion once again is that
technological innovation will provide us a way to make real improvements for
this class of developer.
So, in summary I do not mean to dismiss Linux. It is a serious competitor
which we have to counter with focused development and marketing activities.
Unlike our usual competitors it has a unique economic model, without a
centralized business behind it. In the long run this is a liability, but it can
generate a lot of enthusiasm in the short run.
To counter the Linux threat, we need to focus development efforts on
technological enhancements in the key areas that will matter to customers in the
various segments â particularly case l& 2, but all of them should get some
attention. If we do what we do best â creating and integrating new technology,
weâll pull through OK.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
-----END PGP SIGNATURE-----