OS costs

[Dick Dunn]

In an article following up on looking for good prices on 386ish UNIX,
calhoun at usaos.UUCP (Warren D. Calhoun) writes about finding a good deal on
SCO software...

> ...Considering that even at ~$400.00, there is still more than a
> 100% profit margin (discounting R&D), this is still not a bad deal for SCO or
> for the reseller...

My first reaction to the parenthetical note was to recall physics courses
where we were always searching for the right simplifying assumption...we
joked about "frictionless elephants whose mass can be neglected..."

You can't discount the R&D cost, or the analysis won't make sense.  Now, I
certainly don't mean to flame Mr. Calhoun--it's a common belief that soft-
ware should be very cheap because the production costs are so small.  But
I'd like to take the occasion of his comment to talk about (and perhaps get
a discussion going about) what an OS package really costs, and why.

I think the idea that the OS should be pretty cheap is based on a couple of
points:
	- The hardware is *so* cheap in our world.  If you don't want a
	  window system, a pretty comfortable machine for one or two people
	  costs $2K or under.  A really hot machine is $4-5K.  Moreover,
	  the prices drop weekly...DRAM and disk are as volatile as beef or
	  lettuce.
	- The 386 community is accustomed to thinking high-volume.  There's
	  a sense that everything is commodity-market, here meaning that
	  the price is mostly determined by production costs.

The trouble with this thinking is that the numbers for 386 UNIX systems
don't yet work right for a commodity market.  Obviously, one important
number is quantity; the other one that gets overlooked is time between
releases.  Even if you could neglect your initial R&D cost (which you
can't!), if you're going to stay in business, you have to recoup the cost
of a release during its lifetime--which is on the order of a year or two.
You don't have a cycle where you're intensively designing/implementing/
testing followed by a quiet period of raking in cash.  Instead, the
development is nearly continuous, which means current sales have to support
a full-time development staff as well as the usual maintenance, support,
marketing, sales...  Now, figure that developers cost you $1.5-2 M per
dozen per year...and trust me, you need several dozen developers these
days.  Take a guess at the staff you're supporting, postulate a believable
margin, and you can figure how many systems you've got to sell to stay
afloat.  You might be surprised how high it is.  In vague, hand-wavy terms,
if the margin per system is on the order of a couple hundred bucks, you
need to move > thousands of systems per month.

Another factor, which I find painful but is nonetheless true, is that the
cost of development is monotonically increasing.  Why?  'Cause the size of
the software is monotonically increasing.  With each new release you have
to support 90%+ of the stuff from the previous release, plus all the new
goodies.  (Personally, I think it's unfortunate that we've got such feature
wars going on.  It costs all of us--software cost, performance, relia-
bility--more than we know.)  The continuous flow of new "stuff" makes it
hard to settle down and get a series of stable releases that keep getting
better and cheaper.

BTW, I'm not writing this to cry on anyone's shoulder.  I don't know what
the detailed numbers for ISC are, and I'm not in the Product division any-
way.  These are just the BoE calculations anyone can make.

Can 386 UNIX get to the "commodity market"?  Depends on a lot of stuff.
In the past, we've been on the wrong side of a positive-feedback loop:
Prices are too high, so people don't buy; the quantities stay too low to
get the prices down enough.

Comments?
-- 
Dick Dunn     rcd at ico.isc.com -or- ico!rcd       Boulder, CO   (303)449-2870
   ...I'm not cynical - just experienced.