00100	.group skip 20; fill adjust compact
00200	SDC:
00300	
00400	I learned from Al Blue today that you had not received an expected letter from
00500	John.  Subsequent research revealed that John hadn't sent one and now
00600	doesn't want to for awhile.
00700	Enclosed are a short form version and a longer expression of my views.
00800	I hope that one of these will cover your needs.
00900	.next page
     

00100	.require "good.pub" source_file;
00200	.ltr Mr. Stephen D. Crocker
00300	R & D Program Manager
00400	.arpa
00500	
00600	Subject:  Decommitment of the Processor Project
00700	
00800	Dear Steve:
00900	.fac
01000	In accordance with your request, we have suspended plans to fabricate
01100	a high speed processor.  On this basis, we will not need $200,000 of
01200	the funds allocated under the current contract (SD-183).
01300	
01400	.lde
01500	.every heading(,,)
01600	.next page
     

00100	.ltr Mr. Stephen D. Crocker
00200	R & D Program Manager
00300	.arpa
00400	
00500	Subject:  Decommitment of the Processor Project
00600	
00650	
00700	References:
00800	
00900	1.  "Proposal for Development of a High Speed Processor", to
01000	    Advanced Research Projects Office from Stanford University,
01100	    January 1972.
01200	
01300	2.  Amendment P00023 of Contract SD-183, between Defense Supply
01400	    Service - Washington and Stanford University.
01500	
01600	3.  L. Earnest (Stanford), "High Speed Processor Fabrication",
01700	    letter to E. Stubbs (DSS-W), 12 July 1972.
01800	
01900	4.  E. Stubbs (DSS-W), "Approval of Acquisition or Fabrication
02000	    of Facilities", letter to L. Earnest (Stanford), 11 Sept. 1972.
02100	
02200	
02300	Dear Steve:
02400	.fac
02500	The purpose of this note is to record recent discussions and
02600	decisions regarding our processor development project.
02700	
02800	Funds for this project were requested in January 1972 (Reference 1),
02900	and received in July (Reference 2).  Specific procurement authorization
03000	was requested on 12 July (Reference 3) and approval (Reference 4)
03100	was received at our office on 13 December.  Although we had formal
03200	authority to proceed,  we had requested a design review in mid-January
03300	to insure completeness and adequacy of the detailed design before
03400	certain major procurements were undertaken.
03500	
03600	PROJECT REVIEW
03700	
03800	Our staff went over the design with you and the review committee on
03900	January 15.  The general state of the project at that time was as
04000	follows.
04100	.begin indent 4,8,4
04200	1.  All logic drawings complete (214 drawings).
04300	
04400	2.  16 of 24 printed circuit cards completely designed.
04500	
04600	3.  34 of 44 wirewrap cards completely designed.
04700	
04800	4.  design automation programs operational (and in use at MIT and DEC).
04900	
05000	5.  Prototype printed circuit and wirewrap cards fabricated and tested.
05100	
05200	.end
05300	Our schedule called for two and one-half months more of design review,
05400	paper debugging, and design of the remaining 8 printed circuit and
05500	10 wirewrap cards.  Component procurement was to be partly overlapped
05600	with this phase and all fabrication was to be complete by the end of
05700	June.  Our budget for completing fabrication was $194,315 for
05800	parts and services, and $72,499 for management and staff salaries,
05900	miscellaneous direct costs, and overhead, which totals to $267K.
06000	
06100	Following fabrication, there was to be a debugging period of uncertain
06200	duration, depending on the number and difficulty of problems encountered.
06300	Development of Tenex modifications was to go on concurrently.
06400	
06500	COMMITTEE COMMENTS
06600	
06700	You and the committee were apparently convinced of the technical
06800	adequacy of the design as far as you could probe it in the time
06900	available.  There was some variance of opinion on the schedule,
07000	with the designer's estimates being (predictably) more optimistic
07100	than the committee's.  Even so, there appeared to be a consensus
07200	that the processor could be made fully operational by the end of
07300	this year, barring major catastrophe.
07400	
07500	I heard no criticism of the fabrication budget estimates.  Additional
07600	costs for debugging and initial exploitation of the processor were
07700	not given and are more difficult to estimate.  My estimate is about
07800	$80K.  Apparently, yours was higher.  We were planning to cover these
07900	expenses under our computer facility budget.
08000	
08100	Overall, John McCarthy and I were pleased with the recent performance
08200	of the design group and with the committee's evaluation of their work.
08300	
08400	DISCUSSION
08500	
08600	Subsequent to the project review, you asked us not to build the processor,
08700	citing schedule slippage and a changing technical environment.  Indeed,
08800	there has been substantial slippage in the verbal estimates of the design
08900	staff.  As you are aware, this kind of optimism is common among designers,
09000	especially young ones.
09100	
09200	As far as formal schedule commitments are concerned, we were not so far off.
09300	Our request for approval (Reference 3) estimated completion by 1 February
09400	1973 contingent upon  receipt of approval  by 1 August 1972.  In fact, approval was
09500	received 13 December 1972 and our most recent estimate for fabrication
09550	was about 1 July.  I do
09600	not wish to argue that there was a month-for-month slip associated with
09700	delay in approval, but there was some coupling.
09800	
09900	Of course, a project should not be halted for delays if it still makes
10000	sense technically and there are adequate funds available.  There were
10100	adequate funds available.
10150	
10175	The most recent development on the technical
10200	front, our spies tell us, is that DEC is designing a
10300	machine that may come within a factor of 2 of the performance of the
10400	one we have designed and that theirs will be probably be cheaper than
10500	ours would in production.  That will be an interesting machine if it
10600	pans out.
10700	
10800	Meanwhile, the one that our group has designed is ready to build and
10900	has a much higher performance/cost ratio than anything on the market
11000	or likely to appear in the next two years.  Making the pessimistic
11100	assumption that at most one machine would be built, the question to
11200	ask in our current position is "Is it worth $350K (or whatever figure you believe) to
11300	have a processor 4 times as fast as a KI-10 by the end of this year?"
11400	I believe that the answer is yes.
11500	
11600	DECISIONS
11700	
11800	While not fully understanding your reasons for requesting a halt, we
11900	agreed to suspend procurement on the processor project.  You remarked
12000	that you had no objection to our completing design details and trying
12100	to convince you that the processor should be built, but you assigned
12200	low probability to that outcome.  Our staff subsequently decided to
12300	proceed on that basis.
12400	
12500	You also asked that we inform your office by the next morning of the
12600	amount in the contract that we would not need, given that the processor
12700	will not be built.  I subsequently pointed out that our computer
12800	system remains badly overloaded and that we must get more performance
12900	in some way.  I suggested that the available funds might be diverted to the
13000	procurement of a KI-10 processor from DEC, which would provide some increase in performance and would permit us
13100	to convert to the Tenex monitor.  You said that was not possible.
13200	
13300	On the basis of these decisions, I reported to Al Blue (ARPA) that
13400	we could leave $200,000 unspent.
13500	
13600	CONCLUSIONS
13700	
13800	As I write this, it is 2 AM and there are 22 people running on our
13900	system.  This is not unusual.  Daytime loads almost always exceed 40
14000	jobs.  As far as I know, there is no other PDP-10 installation on the
14100	network that regularly carries half this load.
14200	
14300	As things stand, we have an overloaded timesharing system, a slightly
14400	disillusioned and very disgruntled design group, and little prospect
14500	for improvement in either.  We remain convinced that the new processor was
14600	and is a sound investment and hope to convince you of this.
14700	
14750	.group
14800	In considering alternatives, I trust that you will remember that while
14900	hardware can never compete in performance with paper machines, neither
15000	can paper machines compete with planned machines.  We
15100	solicit your help in finding solutions to our problems.
15200	
15300	.lde
15400	
15500	cc:  L. Roberts, A. Blue (ARPA)