COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	equip.mem[e81,jmc]	SOME CONSIDERATIONS ON COMPUTER EQUIPMENT
C00009 ENDMK
C⊗;
equip.mem[e81,jmc]	SOME CONSIDERATIONS ON COMPUTER EQUIPMENT

	I hope for general agreement on the first group of considerations.
A second group contains opinions which run somewhat counter to
what seems to be the spirit of the times.

	1. We must be realistic about the amount of money we can
expect to get.  The wish list of the previous memo totals more than
$6 million, and Bob Engelmore says we can realistically hope to
get $2 million from DARPA with a possibility of more later.  We
should make our purchase plans taking into account the possibility
that we don't get more later.

	My opinion is that we must run scared about money.  My experience
at M.I.T. when we asked for more money than anyone planned to give
us is that we didn't get a definite negative answer; matters just
dragged on indefinitely.  Therefore, we must maintain contact with
DARPA and other sources after a preliminary proposal.

	2. The proposed file server from Foonly is $205K to $220K,
not $500K as mentioned in the memo.  This includes an F5 processor
at $55K to $65K, a 6250 bpi tape unit at $30K to $35K, 6 disks of
670 megabytes each at $20K totalling 4020 megabytes and $120K.  An
ethernet interface is included in the processor as standard.
Some questions remain about which I have been refraining from
contacting Poole until the dust settles.  These are

(1) Within a year or so, the most cost-effective disks are likely
to be the IBM 3380 or some cheap imitation.  Will the F5 handle
the data rate, and what will it cost for a controller or can
that be included in the processor?

(2) Will the system have adequate error detection and correction?

(3) This one is for us - especially Brian Reid.  Is the F5 fast
enough or would we need an F4?

	I hope this is non-controversial.  Our goal with file
system should be to be able to buy whatever is the most cost-effective
large scale storage.  Already the Foonly proposal will permit
us to give up purges of material prepared by hand.  We will be
able to keep reports and theses on-line indefinitely.  Eventually
we want to be able to keep a library on-line, starting with all
worthwhile computer-represented reports and theses we can lay
our hands on.

	While I haven't mentioned database work, I don't wish to
exclude it from the above.

	3. I see replacing SAIL first and then SCORE.  The
candidates are the 2080 and the S-1, possibly both.
Since neither machine is now available, we need to be sure
that we reserve enough money for one or the other.

	4. The 6 megabuck plan includes $1.5 million for 200
SUN terminals, but there has been a response that it is unrealistic
to expect to get them so cheaply.  If we cannot afford this
much money, then we need a plan for 200 cheaper terminals.  I am
sure we can have bit mapped displays with arbitrary character sets
for a cost of between $1000 and $1500 apiece provided we accept
525 line resolution and use them as terminals only.
I have a detailed proposal for this that needs refurbishment.  An
opus on the use of arbitrary character sets is KEYBOA.PRE[W81,JMC].

	5. Here's the controversial point.  The comparison of
individual workstations and time-sharing has always compared
futuristic versions of the workstations with present time-sharing
in cost-effectiveness.  It was said that an Alto would cost $500
in 5 years, but that was said more than 5 years ago.  The prices
now being charged for LISP machines are realistic estimates of
what a high capacity individual station costs.
Time-sharing isn't dead yet.  Moreover, if we get good terminals
and a good file server on the Ethernet, we can put Lisp crunchers
and number crunchers on the net very cheaply.

	6. Second controversial point.  If some donor gave the
Stanford CSD $6 million to spend as we choose, I think I would
put only about $2.5 million into equipment.  The rest would go
into faculty and space.