00100	                     COMPUTER SCIENCE DEPARTMENT
00200	
00300	                         STANFORD UNIVERSITY
00400	
00500	                     Stanford, California 94305
00600	
00700	
00800	                                                      29 January 1973
00900	Dr. L. G. Roberts
01000	Information Processing Techniques
01100	Defense Advanced Projects Agency
01200	1400 Wilson Blvd.
01300	Arlington, Virginia 20009
01400	
01500	Dear Larry:
01600	
01700		We have decided to complete the construction documentation of
01800	Foonly even under the unfavorable condition that ARPA may  not  allow
01900	it  to  be  constructed  for  non-technical reasons.  Our reasons for
02000	doing this are as follows:
02100	
02200		1.  We really want the machine and will take our  chances  on
02300	permission even though they are not too good.
02400	
02500		2.  We  hope  to  convince  you,  once  the  documentation is
02600	complete, that the machine should be built after all.
02700	
02800		3.  The design aids, which are already being used by DEC  and
02900	MIT will be further proved out by completing the design.
03000	
03100		Admittedly, the machine is not as attractive a proposition as
03200	it would have been had it been  completed  on  any  of  the  previous
03300	schedules.   However, it is still well within the range of completion
03400	times usual for commercial projects.   In  my  opinion,  it  was  not
03500	completed  on  time,  because the group under-estimated how much work
03600	there was and over-estimated how hard each of  them  was  willing  to
03700	work.    It  seems to me that the miscalculations were inclined to be
03800	especially severe in the logical design (now complete) that  requires
03900	sustained  conceptually  difficult  work.    I  expect  that the more
04000	routine  work  of  completing  the  partitioning,   pc-card   layout,
04100	wire-wrap  layout,  and  back panel layout will go much faster, and I
04200	will be interested to see if this is so.
04300	
04400		Unless the documentation is complete by May 1,  I  will  stop
04500	the project myself, but if it is complete, I will try to persuade you
04600	to let the machine be built for the following reasons:
04700	
04800		1. Even one Foonly will provide a cost-effective increase  in
04900	computing  power  for the Stanford Artificial Intelligence Laboratory
05000	and for network users.
05100	
05200		2. The new ideas in the Foonly design such  as  time-sharable
05300	user micro-code and the console computer are worth trying out.  In my
05400	opinion, they better tried out in the Foonly context of an  otherwise
05500	conventional  machine  with  plenty of software, consoles, files, and
05600	users than in an isolated machine devoted to a single new concept.
05700	
05800		3. There is an upper bound on the size project  that  can  be
05900	accomplished  by  a  small group unsupported by a large organization.
06000	The Foonly project appears to have been right on the boundary.  Given
06100	the  software  that  the  project has produced and the development of
06200	technology, this size project should be  well  within  the  practical
06300	range  in  the  future.    It  is important to establish this, and if
06400	Foonly is not completed, there will be a  fear  of  undertaking  such
06500	size projects in the future.
06600	
06700		I  believe  that  Gordon Bell and Alan Kotok will support our
06800	contention that the project is approaching completion, and we  expect
06900	to  have the complete documentation comprising, pc-layouts, wire-wrap
07000	layouts, and back panel wiring ready to be sent out  at  the  end  of
07100	March.
07200	
07300		I hope that when we present this proposition to you, you will
07400	be able to give the matter some of your personal attention.
07500	
07600						Sincerely yours,
07700	
07800	
07900						John McCarthy
08000						Director, Stanford Artificial
08100	Intelligence Laboratory
08200	
08300	
08400	P.S. There are two factors that have unfortunately militated  against
08500	the project at this time, in my opinion, incorrectly.
08600	
08700		The first is the extremely short time estimate given by Poole
08800	and others at the  beginning  of  the  task.    This  was  apparently
08900	necessary  to  him  to persuade himself to undertake the project.  In
09000	fact, the project is  not  badly  delayed  by  commercial  standards,
09100	compare ILLIAC 4.
09200	
09300		The  second  is your and Steve's tendency to push the idea of
09400	concentrating computing facilities in a few  centers  used  over  the
09500	network.  In my opinion, there is a tendency to push this idea beyond
09600	its usefulness as a kind of justification of the expense of expanding
09700	the network.
09800	
09900	
10000	P.P.S. Gordon Bell told  me  that  D.E.C.  is  seriously  considering
10100	contributing to the project the integrated circuits and services that
10200	we had arranged to buy from them.  If I estimate it  correctly,  this
10300	corresponds  to  about $66,000 or about 1/3 of the goods and services
10400	that were to be  purchased.   I  suppose  they  won't  make  a  final
10500	decision until the processor is ready to be manufactured.
10600	
10700		Gordon  also  suggested that we use cheaper ICs for the cache
10800	and microprogram memories which would reduce the  cost  further.   He
10900	said that he hoped that the debugging experience would be written up,
11000	because only IBM and CDC  have  built  pipeline  machines,  and  they
11100	haven't  talked.   I  hope  to be able to accomodate him in this.  He
11200	also agreed to tell you his opinion directly, and I hope he will.
11300	
11400	
11500		Since  Steve  came  out  to  review  the  project,  work  has
11600	continued at a pace that leads me to believe that the April 2 date to
11700	be  ready  for  manufacture is realistic. Details will be supplied on
11800	request.
11850	
11900		In  any case, however, it is not possible to have the machine
12000	all put together by 15 June, so that we hereby decommit this.  I hope
12100	this will satisfy the need for a letter of decommital.