Re: DRAFT Agenda for SLUG/SMBX Meeting of 15 December

[Brad Miller <miller@CS.ROCHESTER.EDU>]

    Date: Mon, 18 Dec 89 16:33 PST
    From: Doug Evans <DE@PHOENIX.SCH.Symbolics.COM>

    Granted, cars are not computers ... but ... how could Symbolics possibly
    stay in business if we came out with a new line of machines using new
    technology and gave them away for a token price and trade-in of an older
    machine?  We can't even buy the RAM memory ICs for a 4MW machine for
    $5K, much less buy the rest of the parts and build the machine.  Who
    else in the computer industry (or any industry, for that matter) does
    this?  Do you have any examples?

I think your analogy breaks down because cars are basically commodities,
that is, their prices bascially reflect marginal cost to produce + overhead
(amortized design, management, etc.) plus a profit. The per-unit profit is
usually pretty small in order to increase volume, and to remain competitive
with other car dealers.

With special purpose computers, one either blows all the money doing limited
production runs (e.g. BBN butterfly, at least in the early days when we got
ours (the first 128 node machine built, btw)), or has an artificially high
markup because the market isn't price driven; you won't sell more unless you
price the machine below the cost of production, so you may as well make a
good profit on the few you do sell.

The point here is, I feel that Symbolics already made their profit off me [*].
Their choice is to provide me with a reasonable upgrade path, or (sooner or
later) lose me as a customer. I bought my machine because it was the best
thing available at the time, AND it was the only machine that met the
requirements of the project. That's no longer the case. Other vendors supply
hardware and software that meet the project requirements for much, much less
money. (I can't name vendors and configurations and prices in the same
breath, but we're talking something that crunches a *lot* faster than a
3620, and costs us less than $20k, plus has the ability to do other things
than run lisp, very efficiently; I may not do too much other than run lisp,
but around here having a machine that can do two jobs well enough is usually
better than having a machine that can only do one, even if it does it very
well).

Other special-purpose machine vendors do the same (with us, anyway, I can't
speak for others). They may make something off us initially, but upgrades
and retrofits are done at cost. (That's in the fairly rare cases they don't
just give us the machines). Examples? Sure. BBN, upgraded our Bflys to +s at
cost (at least, that's my understanding; I don't have the paperwork in front
of me). TI upgraded our explorer-i's to ii's below cost ($5k/machine).  And
they later gave us two free e-i's and let us upgrade them for $5k each.

Other universities (and depts at THIS university) are in the same position
I'm in. So is the situation at large companies (Kodak and Xerox to name two
that are local here and I'm familiar with vis. their lisp environments).

Other vendors (e.g. SUN) do indeed cut us very VERY nice deals (again,
sorry, can't give out numbers publically) in order to assure we don't go
with, say DEC (not to say that DEC doesn't also try to woo us away, but you
get the point). They are already selling machines as commodities, I expect
their price to us generally reflects only the *marginal* cost of producing
the hardware, and they throw in everything else for nothing. From what I'm
told, these cut similar deals with e.g. Xerox; admitably not *quite* as
wonderful, but still *substantially* (30% or more) less than list. And as
you probably know, the list prices on these machines are pretty good to
begin with.

And, indeed, these companies get a lot of benefits from these transactions,
e.g. we graduate people who have used their equipment, which means we
improve the market for the equipment (since companies can now hire people
who know the equipment), and further, some of these folk end up in
management positions and specify the equipment for their own work. The tools
developed at a university today are tomorrows products; look at the good
things that CCI's donation of Tahoe computers to Berkley has done for them
(they gave them Tahoe machines in exchange for their agreeing to port 4.3 to
it, and allowing them to include it in the package they sold to other
customers).

Memory is actually an interesting point; 4mw = 20mb (ok, I'll give you 24mb
for ECC). 70ns fast page rams in SIMMS are $79.; so that much memory is
under 2k. Yeah, there's other overhead, and maybe you need faster rams (?)
5k still seems VERY steep. And indeed the difference between your 10mb and
20mb UXL-400s configurations seems a lot more expensive than 3rd party
memory prices warrent.

Obviously it doesn't work to do this in the case of, say IBM ATs, but here
we're talking about a much smaller installed base, a (presumably) loyaler
customer base, and trying to give them reasons to stay with Symbolics over
other solutions. The Symbolics solution doesn't have to be *cheaper* than
these other solutions, but it must be *cost competitive* with them. Maybe
that means dropping maintainance costs of the older machines, maybe that
means substantial trade-up incentives because the newer machines are cheaper
to maintain... but as Kenneth Forbus <forbus@cs.uiuc.edu> already mentioned,
we can replace our 36xxs with other machines for less than the cost of
maintainance. Symbolics giving us a good reason to upgrade them with new
Symbolics machines, gives us a warm fuzzy feeling about the *next* time this
happens; the maintainance costs climb with the obsolescence of the
technology, and the newer better machines are out. Sun gives us *very* good
reasons each time we upgrade (2s to 3s, now 3s to 4s) to stay with them,
which also bodes well for the next time, that we know will come in just a
couple or three years... will Symbolics do the same?

[*] A particularly wonderful case in point; Kodak wanted to *donate*
Symbolics 3640 machines to us recently; we turned them down; we didn't want
to pay Symbolics $10k licence transfer fees, let alone the annual
maintainance cost of these machines; for each one we took, we could buy 2
workstations from Sun; and this was a *donation*!!