Windows NT (or 2000) vs Linux Comparison ?

Robert G. Brown rgb at
Tue Feb 27 09:36:31 PST 2001

On Tue, 27 Feb 2001, Yoon Jae Ho wrote:

> Can you help me to find out the comparison information between Windows
> NT (or 2000) vs Linux Comparison ?

I'm not sure what you mean (and this is perhaps mildly off topic,
although integration of beowulfs and linux clusters in general into
WinXX dominated environments isn't completely irrelevant to the bigger
picture, if not particularly technical), but let's give it a shot.
Beowulf technoids can always hit the old "d" button about here --> <d>


Costs (typically) $100 to $3700 retail (the latter for Win2K Advanced
Server with 25 client access licenses).  Upgrade costs slightly lower,
but only slightly.  Historically "requires" an upgrade to maintain
compatibility every 2-3 years, although a few people managed to stretch
e.g. Win 3.x or Win95 out a bit longer.  Required office suite (Office,
of course) costs $190 to $800 retail, depending on the included
components.  Compiler/development suites, not included, cost $500 to
$900 retail (if you stick to Microsoft products, perhaps $200 and up if
you go with other companies). Finally, there are applications packages
for all sorts of things -- Netware, virus protection (pretty much
required on any WinXX system) backup packages, desktop publishing,
database operations -- that cost anywhere from $30 to $1000 depending on
what you get and how current it is.  Let's say that a minimal desktop
configuration (Win 98 or ME) including the cheapest version of office
(which omits useful things but so it goes) and an antivirus package but
NOT compilers and so forth is perhaps $300 retail.  A high end
developer-class workstation configuration might cost $2500 or more for
software alone, full retail, or obviously as much more as one needs to
spend for particular software packages.  Client/server configurations
look like they'd average at least $500 a seat, not including compilers
or a lot of extra server.

Microsoft has created a gladsome world where they are likely to make
far more actual profit per computer sold than the retail seller of that

Additional costs not included in the above but critical in any wide
scale adoption of WinXX on an enterprise basis include the non-scalable
management costs of administering what is essentially a GUI driven
operating system interface on a large collection of networked boxes.
This limits, fairly sharply, the number of systems a single
administrator can install and manage.  Humans trained and certified in
systems work tend to be expensive.  I'm going to fairly arbitrarily
judge that a single $60K person can manage 60 Microsoft desktops in a
typical organizational setting, which adds $1000 apiece to their cost.
Note that this number can vary wildly from organization to organization,
depending on many things, so it will almost certainly need to be
adjusted.  Some places with unusually skilled people might scale up to
several hundred seats per administrator.  Other places only have ten or
twenty seats in the first place, but still have to get an MCSE to care
for them full time.  Overall, I cannot see admin costing less than
$300/seat/year and in most cases it will be many times that.  Folks
familiar with outsourced Windows admin costs (probably a fair basis for
low end costs for large organizations) should feel free to correct me.

To be fair, a large organization or University can almost always reduce
per seat costs of Windows software by dickering out a site license.
Duke, for example, pays only a small fraction of the non-human costs
indicated above per seat.

Benefits of using Windows include an extremely wide range of commercial
software to choose from, lots of great game software to distract your
kids and employees and self from constructive tasks, the widest possible
range of hardware device support, and a simple, intuitive interface to
most software that holds down training costs and user support costs.  A
backhanded "benefit" is that Microsoft has made e.g. Office formats a de
facto standard in many environments, which ensures "intercompatibility"
for any organization that chooses to standardize on one particular
flavor of WinXX and Office and so forth.  Finally, since the needs of
90% or more computer users are met by anything that allows them to
create and edit formatted text documents, send and receive mail, browse
the web, and possibly run simple spreadsheets, a standard
office-equipped PC (the $300 software cost system mentioned above) is
indeed an adequate desktop solution for most users and can be "managed"
by complete computer-idiots from its GUI in a home or very small office
environment, with varying degrees of success (but success nonetheless).

Antibenefits of using windows (exclusive of the out of pocket costs
discussed above) include reportedly lower stability -- I can vouch for
this personally up through 98 but haven't managed to get a dual boot
Win2000 box up yet.  Part of this instability is associated with a
significant and well-publicized vulnerability to highly destructive
computer viruses that can be imported by careless users both in
applications and in e.g. email attachments where it is difficult to
guard against them.

Windows also suffers tremendously from a lack of NON-GUI based systems
interfaces and proper multiuser/multitasking capabilities in 9x, and a
lack of secure remote network accessibility in all flavors.  This latter
disability has many disadvantages.  Itemizing, it is extremely difficult
to build a network-distributed scalable management scheme.  It is
difficult to impossible to login to your WinXX system to do or check on
work in progress when not at the console of that system.  It is
difficult to impossible to open graphical tools running on system A and
display them and manage them from system B.  Overall, Windows is stuck
in a time warp of network obsolescence at least seven or eight years

I think it is this that annoys me the most about WinXX (he says as he
sits at home, typing on linux box A, piping keystrokes through a
masqueraded connection through gateway linux box B, sent to a mail
program running in the department on linux desktop C, while running a
remote GUI that allows me to monitor job progress and CPU activity on a
small cluster of remote systems, all without any real hassle associated
with it).  It is incredibly limiting to one's productivity to be chained
to a particular desktop and has been for years, yet WinXX products
persist in perpetuating the one user, one seat "personal computer"
paradigm of the 80's.  The separation of server from client capabilities
also falls very much in this category as a disadvantage -- one has to
basically reinstall a system with a new, much more expensive version of
the operating system in order to convert a "client" into a "server".

Windows itself is delivered in a nearly useless base configuration -- it
only becomes useful as additional software is purchased, starting with
some sort of office suite, since in its baseline configuration it
doesn't really support all of the basic user tasks indicated above.
There are expensive transitional points between versions of WinXX and
mixed-function applications.

A significant antibenefit of Office in particular is that although it
presents itself as an "intercompatibility standard" it is in reality
anything but.  Documents produced by office components have historically
been highly NON portable -- often backwards incompatible with Office
itself, occasionally incompatible with Office itself running on
non-Microsoft platform. <wry>For some mysterious reason, users of
Microsoft software components find themselves inevitably channeled
towards the adoption of WinXX in its latest incarnation running Office
in its latest incarnation.  Of course, as Microsoft itself is fond of
saying, they aren't anti-competition or anything...;-)</wry>

Then there is the overwhelming barrier between Office users and
everybody else -- most of the world interprets the phrase "document
standard" to mean "open, well documented, publically agreed upon" so
that anybody can write a document manipulation tool to open, edit and
save documents of that type.  RTF, for example, is such a standard for
formatted text.  HTML and SGML also represent standards of this sort.
Users who work in environments where real open standards are the
non-competition-stifling rule get extremely annoyed when documents are
sent to them in a proprietary format, and this can form a real (and
expensive!) barrier to interaction outside a WinXX/Office-version
homogeneous domain.  Similar barriers hold, of course, for other
proprietary document formats produced by other commercial software
vendors used on WinXX platforms.

A final, extremely important antibenefit of Windows is the lack of
well-documented access (or even UNdocumented access) to signficant parts
of the system -- no access to software sources, no access to the guts of
the operating system or even its functional system specific interface,
no access to its device drivers -- except as provided by an extremely
limited GUI.  If something works via the provided GUI interface out of
the box you are golden.  If something doesn't work, you're sunk and it's
nobody's "fault" (that is, there is nobody you can hold legally or
financially responsible for fixing it) so you're going to end up paying
for it one way or another.

Note that this is usually sold as a benefit -- all things on a Windows
system are supposed to just work, and work simply through a predictable
and intuitive interface.  Reality, of course, often dictates otherwise
and the frustration and lost productivity associated with trying to
"fix" the essentially unknowable and unfixable within the confines of a
finite set of permutations of GUI buttons in my opinion at least easily
balances the benefits realized when it works, especially when one
realizes that this needn't be an either/or design choice -- one can
easily engineer and document things to be manageable by both GUI (when
it works) and the hard way (but still much easier than no way at all)
when they don't.

To summarize, WinXX on an organizational scale is quite expensive --
probably averaging $600-1000 per seat at the time of installation and
with an incremental cost of some $500-700 per seat per year (including
both amortized revolving software upgrades and scaled management costs).
Its primary benefits are ease of use and minimal training costs for the
most commonly used user software and an internal intercompatibility
realized for any organization that adopts it exclusively, the widest
range of commercially available software, and numerous avenues for
obtaining paid support.  Its primary antibenefits include lack of
stability, lack of network accessibility and network-based scalable
management options, a closed source, poorly documented, mostly hidden
kernel interface, and significant incompatibility/interoperability
problems in a heterogeneous environment.


Costs (typically) $0 (well, pennies per seat) to about $150 for a
box-set professional server edition.  Note that there exist linux box
sets that costs as much as $2000, but these are for high-availibility
clustering servers intended to run e.g. massive distributed websites
(these are not beowulf-style compute clusters, for sure) and aren't
really something one can compare to anything in the WinXX lists above.

Amusingly, for the $0 cost (really anywhere from a few dollars for cheap
CD's or the net access required to download a free distribution to that
full $150, amortized across the number of seats one wishes to install
upon), one gets the operating system, a full suite of compilers and
development tools of all flavors, a choice of GUI's, a selection of
office tools that include web browsers, WYSIWYG formatted document
editors, mail tools, spreadsheets, and a choice of full relational
databases, all plural by intention, as well as all the software required
to make the desktop host in question a server, a client, or any mix in
between of the two.  No licenses are required for any part of the
package.  All sources to everything in all the packages is included,
generally in a prepackaged build-ready format so one has BOTH the
sources AND the ability to make simple changes and immediately and
consistently reinstall or upgrade the applications from the modified
sources.  A single CD, or a two CD set, suffices to install an entire
department with any mix of clients, servers, hosts that function as
both, hosts that function as gateways/firewalls, hosts that function as
beowulf or HPC cluster nodes.  It is rare to require any additional
software, but commercial office suites and database products and
compilers are certainly available for prices ranging from a few dollars
a seat to a few hundred dollars a seat, and some task specific
commercial tools that might cost even more (on any platform).

Still, the average retail software cost per desktop in any organization
has to be guestimated as being in the $30-50 range per seat, and will
only be that high if the organization feels an obligation to actually
buy box sets of the distribution selected for every few seats or spends
money on third party documentation (which might just as easily be needed
and purchased for a similar WinXX installation).

The per seat costs for management must also be considered far lower for
linux than for Windows, with some caveats.  First, there is no question
that linux distributed systems management scales far, far better than
WinXX ever could dream of scaling with its current architecture.  It is
straightforward to set up an organizational network where any box can be
installed with linux over the network at zero marginal cost to the user
from a single common source repository using a single, downloadable
floppy image (we have such an arrangement at Duke -- I've used it and
know that it works quite well indeed).  Within a department, tools like
kickstart can be used to almost totally automate installations and major
upgrades, other tools (some still under development) can totally
automate security updates and intermediary upgrades.  One skilled
systems manager, armed with these tools, can now manage a network of
moderately homogeneous linux hosts of truly awesome proportions.  A
network of hundreds of hosts is fairly easy.  The limiting factors are
much more the number and complexity of the servers required and the need
to provide user support to folks at all of those seats than anything
associated with systems installation and maintenance, in stark contrast
to WinXX departmental LANs where installing a system requires order of
hours of time and where arranging an upgrade of a hundred-host
department requires either a team of trained installers or days of
downtime (or "mixed" uptime).  Our linux sysadmin has managed minor
upgrades of some 150 hosts in a few hours work, mostly spent developing
a single distributed script to be triggered by cron.

A fair estimate of cost (comparable to the one above) might have a
single $60K admin caring for a network of 180 hosts (mostly clients, but
with 5-10 servers of various sorts as well.  Again these are real-world
estimates based on what we have accomplished at Duke; your mileage may
vary in either direction.  A small organization might not have 180 hosts
of any flavor to support; a large organization might find that it could
run 1000 desktops with twenty or thirty servers with a team of two or
three (depending mostly on how much training and user support is
required -- two admins could support 2000 boxes if no user support is
required, assuming rollover replacement every three years (or roughly
two new systems a day), onsite hardware service contracts, moderate
control over hardware configuration (sorry, no exotic or unsupported
hardware) and perhaps fifty servers, and nothing fancy required on the
side -- these admins would NOT be your webmasters, java developers,
WinXX problem debuggers, hand holders.

Another significant administrative cost-lowering advantage of linux in a
complex environment is that with a well-documented, primarily ASCII file
interface to the kernel and all major operational tools, with instant
access to compilers and open, freely modifiable sources for all of those
tools, and with a wide range of networked support interfaces (mailing
lists, websites, development repositories like sourceforge) a skilled
systems programmer can debug and fix nearly any quirk that arises in a
given network installation.  By this I mean that showstoppers in a Winxx
environment, like a complete lack of some critical service or daemon,
become mere bumps in the road under linux.  In an environment that does
NOT stretch administrative resources to the theoretical limit (generally
a bad idea in any event) the opportunity cost time liberated during
periods when the network is stable and kept functioning with little
effort on the part of the administrators the creative talents of those
generally talented programmers is indeed liberated by complete access to
the full palette of systems development libraries and tools and your
admin group can indeed also be webmasters, code developers and so forth
and even find some time on the side to debug problems over in the Winxx
side if there is one or hold hands here and there with terrified users.

This also accurately describes our local environment -- our primary
sysadmin can manage his 150 and rapidly rising linux clients, our
central servers, and a crop of heterogeneous and crochety research group
linux mixed client/servers and still find adequate time to mess with a
limited number of NT systems in the labs and front offices, hold the
hands of distraught users of all sorts, contribute to the NFS drivers in
the current linux kernels, and even have time leftover to help manage
our LUG site, including the aforementioned have-floppy, can-install 7.0
ready-to-run FTP drop.  Seth facilitates the installs and upgrades of
literally hundreds of linux boxes all across campus (in addition to
fully managing our department) and has enough time to spare to do other

Again to be fair, linux still does have a more limited supported device
list than WinXX, and systems persons will waste more time getting
marginally supported devices to work.  If an organization is willing to
force all users to use Office (and pay the up front costs of doing so)
one would imagine that it would be willing to pay the relatively small
cost of creating and certifying an approved hardware list.  This is
something that most organizations would be well advised to do even with
a Winxx base, after all, both to achieve the greatest economy of scale
on hardware purchases and because Winxx itself is not immune to hardware
installation glitches and broken or incompatible device drivers, all of
which suck up administrative time and reduce stability.

It is also true that linux is an Unixoid OS and hence "expert friendly".
It is much more difficult for novices to learn to manage linux in even a
minimal install configuration than it is for them to deal with Win9X.
Users who attempt to do this using only the provided GUI tools not
infrequently encounter barriers where things don't work using the GUI
only to manage.  OTOH, those users have some hope, with linux, of being
able to solve their problem eventually outside the GUI. It should also
be noted that WinNT or Win2K boxes aren't much easier to manage if run
as moderately secure, multiuser networked platforms.  A lot of the
difficulty comes from linux NOT being a "personal computer" interface,
and the additional layers of functionality inevitably bring additional
complexity on ANY OS platform.

Benefits of using linux throughout an organization (from node to desktop
to server) include lowest possible cost in software, lowest possible
cost in scalable human management, greatest possible interoperability
and compatibility and standards compliance, superb networking, superb
GUI with full ability to execute GUI-moderated tasks remotely as easily
as locally, full/continuous zero-cost path between base/client
installation and Internet-wide server installation, ability to recycle
unused desktop CPU cycles into numerical or other tasks, ability to
build COTS HPC supercomputers (beowulfs) or other forms of HP compute
clusters (NOWs or COWs), and a far more symmetric and powerful user
installation, where users on ALL desktops can access the full suite of
code development tools, office productivity tools, general application
tools, multimedia tools, and web tools with no additional marginal costs
or maintenance efforts.

Antibenefits include a significantly higher risk of network-based
security exploits that cost one downtime or compromised data (risks
somewhat mitigated by the total impossibility of "virus" like attacks
that recently have been far more widespread and costly), significantly
greater skills required to achieve true admin or systems programming
mastery (linux admins -- rightly -- scoff at most "MCSE"s) and a
consequently somewhat more limited hiring pool from which to draw
reliable talent, considerable difficulty importing MS/Office generated
documents and manipulating them in native form (a difficulty that could
only be truly mitigated by the breakup of the Microsoft monopoly or the
generalized rejection in the marketplace of a non-open "standard" for
key document components -- both are possible and the latter even
probable over the medium term).  Less commercial software is available,
although this is largely offset by the large base of functionally
equivalent open source software included in the basic distributions.
Fewer devices are supported, although the "device gap" is considerably
smaller now than it was one, two, or three years ago.  Finally, although
linux now has excellent GUIs and GUI tools in both the Gnome and KDE
suites (with more on the way) a lack of user familiarity with these
interfaces and tools is a small but niggling barrier to their adoption
over Winxx tools and interfaces among users who are familiar with and
used to the latter.

In summary, linux is cheap, scalable, powerful and functionally
complete, at the expense of greater complexity and hence a greater level
of personal expertise required to get the most out of all that power.
It lags Winxx in device support and games (rats!) and certain (mostly
commercial) tools but the gap is rapidly closing and should continue to
close quite inexorably over time.

In comparison, I think that the Winxx vs Linux choice is a no-brainer.
Where linux is sufficient to meet the software needs of the users, it is
the overwhelming cost-benefit-optimizing choice for an organization to
adopt from the desktop through their largest server racks or dedicated
supercomputer cluster.  In environments where software must be run that
is not available for linux (for example in our physics student labs,
where the lab software that runs with the transducers in the lab kits is
WinXX based) one should regretfully run e.g. WinNT or 2K and pressure
the software manufacturer for a Linux version ASAP (while looking for
commercial or open source linux alternatives).  Organizations that
expect to EVER communicate externally should avoid endorsing Office as
any sort of de facto standard, and indeed should likely set up true
interoperable standards for their critical documents (e.g. RTF, PDF) or
(in a pinch) even those Office formats that import and export adequately
in competing office suites to facilitate true document portability.

This will provide those organizations with the greatest possible
flexibility and opportunity to minimize costs -- if linux continues to
improve (and I'm certain that it will because what else can it do?)
functionality can be shifted to it smoothly at near-zero marginal cost
as it becomes cost-beneficial to do so (as each successive
mission-critical application becomes stably available for linux).  Ditto
for scaling up a linux administrative group -- once it is established
enough to provide good administrative scaling and possesses enough local
linux expertise to support the local degree of complexity, it can be
expanded within the organization at a NEGATIVE marginal cost.  After a
brief period of investment, linux will SAVE the organization money for
every system that converts from Winxx.

In the long run, this is all that matters.  In the long run, this is why
Linux will "win".  Lower marginal cost at equal or better productivity
is money in the pocket, and money in the pocket is profit or investment
capital.  Linux has been growing at the highest possible rates in server
markets for all of the reasons above.  It is also growing at a high rate
on the desktop as more folks make it over the initial hump associated
with the conversion from Winxx and as the wisdom of setting up an
integrated/transitional Winxx and linux becomes more apparent to
corporate managers who've seen it work in their server shops.  With the
minimum cost per seat of Winxx in any flavor at easily hundreds of
dollars, a minimum cost per seat of tens of dollars or even less is
infinitely seductive, once experimentation and success stories published
in business and computing journals make it clear that the decision to at
least enable a gradual movement toward linux isn't as expensive or risky
or irrevocable as Microsoft would like you to think that it is.

Note that eventual success of the linux "strategy" is hard for
Microsoft, or any other non-open source competitor (e.g. freeBSD) to
prevent.  Microsoft can lower their prices (they have PLENTY of room to
lower their prices) but since they've already saturated the market,
lowering prices comes directly out of their profit margin and hence
directly-without-passing-go out of their stock prices, which are scaled
at P/E ratios appropriate for strong growth companies, not companies
that have saturated a market and are suffering gradual erosion of market
share to a competitor that can get by on the lowest possible marginal

What would they have to do to "compete"?  Bundle the operating system,
Office, their compilers, full server support, basically the whole nine
yards of user software into a single installation and then sell it for
less than $100.  Quite aside from the antitrust screams this would
produce in the software world from dying software companies that live in
the grey area between Microsoft prices and linux (content to live on the
meat that clings to the bones after Microsoft has picked them nearly
clean) this would cost Microsoft 80% or more of their revenue -- they
might make a couple billion dollars a year this way, but not twenty.

Sooner or later, regardless of what the courts do now with Microsoft,
they will be forced to bundle most of what they sell piecewise at far
greater profit now and face the Scylla and Charbydis of dramatically
lowered profits AND an antitrust suit they will almost certainly lose.
This isn't an opinion -- it is pure prophecy based on a knowledge of one
of the fundamental forces of nature -- the almighty dollar.  They will
go down kicking and screaming and trying to invoke antitrust suits on
companies like Red Hat that don't own any part of what they sell -- and
they will lose on both constitutional grounds and because there is
nothing to prevent Microsoft from selling stuff THEY don't own to

Exxon would have a hard time arguing that folks selling solar cells made
from a public domain recipe involving aluminum foil and toothpaste that
can clearly be made at home or made and sold by anyone who wishes is
competing "unfairly" in the realm of energy sources in any meaning of
the word unfair but "it's unfair that you take my extremely lucrative
and otherwise unassailable monopoly away from me by charging low prices
for energy that is freely available".  I'm curious to hear how their
latest gambit -- the "it stifles innovation for you to freely give away
YOUR innovations and put our extremely lucrative and otherwise
unassailable monopoly out of business so that we can no longer sell OUR
innovations" ploy -- will play in the humourously decorated halls of
high power.

Enough.  Hopefully somewhere above you will find whatever arguments you
need to move your organization(s) into a slightly saner cost/benefit
strategy based on real (i.e. portable, common, open) standards.


Robert G. Brown	             
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567  Fax: 919-660-2525     email:rgb at

More information about the Beowulf mailing list