From: utzoo!decvax!cca!CAULKINS@USC-ECL@sri-unix
Newsgroups: net.works
Title: CPU-per-user architectures
Article-I.D.: sri-unix.1477
Posted: Sun May 16 22:52:22 1982
Received: Mon May 17 04:20:24 1982

Here's a series of exchanges concerning my suggestion of CPU-per-user
architectures in WORKS Digest V2 #151.  My apologies for repeating
some stuff from earlier digests, but I don't believe the trains of 
thought would be clear otherwise.

Date:  5 May 1982 2250-PDT
From: CAULKINS at USC-ECL
Subject: Re: WORKS Digest V2 #51
To:   Deutsch at PARC-MAXC
cc:   CAULKINS

In response to your message sent  4 May 1982 16:42 PDT

Any interactive workstation is going to need "case, keyboard, CRT,
and other non-computer elements".  So the savings is strictly in (1) the extra
packaging not required for the processors and memories, and (2) the sharing of
the disk.  I don't see that making the difference between "less than $2K" and
$5K.  What did I miss?

[Several points:

1) The manufacturing wizards at my company tell me that the cost of all non-
silicon parts of a workstation type device (case, power supply, et al)
is 30 - 50% of the total cost of the device.  These things are not rapidly
cost-declining technology like the silicon parts.

2) The workstation folks have not gotten nearly as far up on the economy
of scale curves as the terminal manufacturers - I can get a very respectable
terminal for $450 today; about 10% of the $5K workstation price.  Until
and unless a lot of shaking out takes place the workstation manufacturers
are unlikely to achieve terminal-like scale economies.

3) Two important workstation/CPU-per-user architecture parts - hard disks
and power supplies - have decidedly convex cost-per-unit-resource curves;
$/amp and $/megabyte decrease markedly as the capacity of the PS or disk
drive goes up.  The CPU-per-user wins here because many users share 
bigger disks and power supplies.

Conclusion - cost/effectiveness favors an architecture taking full advantage
of economies of scale in terminals, disks, and power supplies; and one
with a maximum amount of value in silicon as opposed to plastic, iron,
and copper.

Dave C]
-------
Date: 6 May 1982 09:20 PDT
From: Deutsch at PARC-MAXC
Subject: Re: WORKS Digest V2 #51
In-reply-to: Your message of 5 May 1982 2250-PDT
To: CAULKINS at USC-ECL

Hm.  What you say about workstations and terminals may be true for the current
generation of character-display terminals.  It certainly isn't true (at least not yet)
for bitmap displays.  The price of $450 for a character-display terminal
(including an interface to some kind of shared bus, Ethernet, or serial line)
seems low to me, and the price of $5K for a workstation is much too high if it
doesn't include a disk.  I.e. I think you need to distinguish the disk-sharing part
of your argument (which I basically agree with) from the power-supply part
(which I am less sure about) and the processor/memory part (which I think is
more complex than you make it out to be).


Date: 6 May 1982 09:24 PDT
From: Deutsch at PARC-MAXC
Subject: P.S. Re: WORKS Digest V2 #51
In-reply-to: Your message of 5 May 1982 2250-PDT
To: CAULKINS at USC-ECL

It seems to me that your statement that "the cost of all non- silicon
parts of a workstation type device (case, power supply, et al) is 30 -
50% of the total cost of the device.  These things are not rapidly
cost-declining technology like the silicon parts." actually suggests
the opposite conclusion from the one you want to draw.  If the case,
keyboard, and CRT cost essentially the same for a workstation or a
terminal (which they do) and they represent a growing fraction of the
device cost (which you assert they do, i.e.  their cost is declining
slower than that of memory and logic), then as time goes on it becomes
LESS important to share those parts of the device (the processor and
memory) whose cost is dropping.

16 May 82

I (Dave Caulkins) finally get some time to continue this discussion:

One thing I think we've learned about the economics of modest (under
$5,000) computing equipment is that it is very price-elastic, and in
a non-linear way.  The success of Apple, Sinclair, Tandy, etc. seems
to indicate that when the price for a particular unit of computational
capability is lowered by N% (where N is 10 - 50), the market for this
particular unit grows by much more than N%.  This suggests to me that
it makes more sense (from a marketing point of view) to lower cost than
it does to expand capability, given that a particular computational
capability can be made available for a lower cost.

I'll apply some crude cost rules-of-thumb to illustrate:

Assumptions:

1) The ratio between the packaging (case/keyboard/CRT/power supply)
cost and the total cost (the packaging plus the electronic (ICs/PC
board/disk) parts) of a workstation (WS) type device is 0.4; i.e. (total
cost)*0.4 ~= packaging cost

2) The retail price of a workstation type device ~= 4 *(total cost).

3) A reasonable non-bit-mapped terminal has a current retail price
of $500, quantity 1.  My entire case is based on non-bit-mapped
systems; it is not clear to me that bit-mapped displays contribute
that much to the value of a workstation.  This is, of course, a
highly subjective judgement.

Using these numbers and a workstation price of $5000, then:

5000/4 = 1250, WS cost

1250 * 0.4 = 500, WS packaging cost

1250 - 500 = 750, WS electronics cost

750 * 4 = 3000, WS electronics price

3000 + 500 = $3500, WS electronics plus terminal price, the approximate
price per user for a CPU-per-user architecture.  This represents a 30% 
savings over the ordinary WS price, quite substantial enough to 
induce the non-linear price elasticity mentioned above.
-------