[OPINIONATED AND LENGTHY TREATISE FOLLOWS - BLAST SHIELDS UP!]
Tony -
Thank you for giving us the opportunity to provide you with feedback. I've
read your attached paper and I thought I'd send you some comments (cc'ed to
the ALE list). My perspective is that of an IT manager who has analyzed TCO
issues under Intel+VAX and Intel+Novell/Banyan/NT environments and who has
been working with Linux for about two years overall.
In your paper, you say that "Since operating systems are usually bundled
with hardware, Linux users don't save on up front OS licensing costs." This
is the only thing in the paper I seriously take issue with you on. Windows
(98/NT/2000) bundling is the norm for purchases of individual machines from
Gateway, Dell, Compaq, etc. but even so, the consumer *is paying* even if
there is no specific line item identifiable as such. If you use any of the
"autospeccers" on Gateway's, etc. Web site, if presented with a choice of
OS, you will see an added cost (maybe $150 or so?) for NT Workstation. Just
because there is no dollar amount attached to the default choice of Win98
DOES NOT mean that the consumer isn't paying.
IMHO, only the most derelict IT departments would actually buy PCs this way
- from a megavendor with OS preinstalled, where you'd make a few network
settings changes, install a few apps, and hook it up to the network. In my
experience, it's more cost-effective and flexible to decouple machines and
OS licenses, i.e., you buy OS-less PCs in quantity, you buy a certain number
of seats' worth of OS, design your own OS distribution to run on the PCs,
and beam it out to new PCs as they are installed. Operate this way and you
won't have your OS costs buried within the cost of new PCs.
In "Win-world" your OS costs go as follows:
TCO(OS) = AS + (B + C)W
where
TCO(OS) = Operating system component of Total Cost of Ownership
S = number of servers
W = number of workstations
A = cost of server OS license
B = cost of client OS license
C = cost of client access license for server OS (presuming MS' "per
seat" licensing)
What I've done in the past is to make a carpet plot of TCO(OS) as a function
of any two variables (S and W are usually the most meaningful, treating A,
B, and C as constants). When doing similar plots for other OS mixes (Banyan
and Novell, for instance), sometimes there will actually be discontinuities
in the plot because of the vendor's pricing structure for a given product,
thereby making one or more of A, B, or C non-constant. When comparing
different combinations of different products, there will sometimes be places
where one combination undercuts another, often as a result of those
discontinuities.
But anyway, I digress. The most important thing to keep in mind about Linux
as it relates to TCO(OS) is that with Linux as both client and server OS, A,
B, and C are all effectively zero. This is why I often use the phrase
"Linux costs don't scale." Put another way, whereas using Linux may affect
an organization's overall IT TCO in a large number of ways, the sheer number
of servers and workstations you have isn't one of them. If you decide to
build an environment with Linux servers and Win98 clients, your A and your C
will remain essentially zero. In our organization, we have 10 NT servers
and something like 70 mostly-Win98 clients. I'd estimate our Win-world OS
costs as (10 * $600) + (($80 + $30) * 70) = $13,700. If we had done Linux
servers instead of NT servers, we'd have cut that by a little under half.
In real-world situations, I think one would find that if you live in
Win-world, you're always having to buy "fiddly-bits" to accomplish small but
important functions. X servers. NFS clients. Time synchronization.
Defraggers. Backup software. Image-editing software. Compilers (you
already touched on this in your paper and approriately so). In Linux-world,
these kinds of things are either not necessary, already present, or sitting
an Internet download away.
After you've worked with Linux for a while, you begin to expect the tools
you need to already be sitting in front of you, so much so that it becomes
second nature. NT doesn't come with as much as a FORTRAN compiler; if you
want to develop software under NT, you have to choose your
language/environment/etc. carefully because you have to pay to do any ONE
thing - and if you decide to do something else, you have to pay for THAT one
too and the money you spent before on the PREVIOUS thing *stops giving you
value* (that's a fancier and more accurate way of saying "is wasted").
There's something you're unlikely to be able to actually quantify TCO-wise.
More so than any OS I've ever worked with, LINUX REWARDS HARD WORK,
INITIATIVE, CREATIVITY, ETC. If I had to oversimplify and overgeneralize,
people who work only with NT (I'm speaking as someone who was once regarded
as the chief proponent, advocate, and implementor of NT in two different
organizations, so don't peg me as an NT-basher) can only take things so far,
whereas people who work primarily with Linux (and/or, dare I say it, other
Unix variants/clones) are more open-ended, by which I mean that as time goes
on and new challenges arise, the *ux person is more likely to learn some new
"kung fu" to handle them. One the other hand, I've seen a number of NT
people kind of stagnate - they can't see a path forward on a problem (not
necessarily wholly their fault, mind you; paths may not be there to see), so
they just don't make any progress against it. I would much rather have
people working for me who were accustomed to and got a lot of practice in
solving problems.
I personally get a big kick out of running Linux effectively on what other
people regard as uselessly dated hardware. Since it's quite do-able to do
remote centralized computing a la Windows NT Terminal Server Edition in a
Linux client-server environment, I could go into an organization that would
otherwise plan a full replacement of hundreds of 486 and Pentium client
machines in order to do an NT client/server app deployment and say, "keep
'em - say hello to your new X terminals!" I could save a company hundreds
of thousands of dollars that way.
- Jeff
--
To unsubscribe: mail ">majordomo@ale.org with "unsubscribe ale" in message body.