COMMENT ⊗   VALID 00011 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	
C00006 00003	
C00009 00004	
C00012 00005	
C00015 00006	
C00018 00007	
C00019 00008	
C00020 00009	
C00022 00010	
C00023 00011	
C00026 ENDMK
C⊗;






				     DRAFT





			    Displays for the Masses





	   The goal was to prepare a plan for Project MAC and the AI Lab to

   acquire a system of keyboard and display terminals large enough and cheap

   enough to put one in every office in 545 Tech Square - about 130

   terminals.  The system is to be connected to three PDP-10's on the ninth

   floor, and to MULTICS in building 39.  Desirability of a direct

   connection with the ARPA net was proposed later.  The performance goal

   was 20 or more  80 character lines of arbitrary characters plus graphics.

	   The committee to prepare this plan consisted of Tom Knight, John

   McCarthy(chairman), and Jerry Saltzer.  It was assisted by others from

   MAC and AI.  In addition, we believe we have consulted all interested

   groups in MAC/AI.

	   An RFP (request for proposals) was prepared and sent to Data

   Disc, Ramtek, and Owens-Illinois with a reply date of 1 November for

   proposals.   Proposals have been received from Data Disc and Ramtek, but

   Owens-Illinois requested and received an extension until 15 November.

   RFP's were also requested and sent to DEC, Evans and Sutherland, Systems

   Concepts, and Imlac.  Of these, Systems Concepts is known to be

   interested  in submitting a proposal.  The others have not attempted to

   arrange the meeting mentioned in the RFP.







							Displays      Page 2



	   After the RFP's were sent, we learned about the approach taken by

   Peter Weiner of Yale.  We visited them, and now recommend a slightly

   modified version of their approach.  Their approach is based on the fact

   that complete TTL interfaced semiconductor memories of better than 450

   ns. cycle, organized into blocks of 4x4k x 16 words can be obtained for

   .86 cents per bit from at least two suppliers, AMS and Intel.  Since our

   purchase will be four times as large, we may get a better price.

	   This allows storing a 512 by 512 dot matrix raster for less than

   $2200, without the maintenance problems and unexpandability of the Data

   Disc system. It beats Ramtek by a factor of two in cost, and is much more

   flexible than either, because the memory is directly and randomly

   accessible to a mini-computer.





	   The proposed system consists of the following:



   1.  A PDP-11/45 computer to control the system.



   2.  Interfaces from the PDP-11 to the three PDP-10's, MULTICS and later

   an imp or tip.  Connection to the 370/165 run by IPC is also possible.



   3.  1024K  16 bit words of IC memory. The Intel version  bought by Yale

   has 4k words on a card and for each four such cards, a control card, such

   that 16k words form an independently addressible unit, producing up to

   four 16 bit words in parallel.  Our system would contain 64 such units,







							Displays      Page 3



   so that 64 terminals could be active simultaneously. 



   4.  A two port memory control / shift register card per console.  One of

   the ports reads out into a 64 bit shift register under control of a

   centralized video timing generator.  A 512 by 512 raster is presently

   contemplated, but this is easily changed.  The other port interfaces to a

   large address space buss, driven by, among other things, the PDP-11.

   Half of the PDP-11's 32k memory space refers to a selected display's 16k

   words.



   5.  The 64 video signals go to a video crossbar switch which is presently

   envisioned having 72 input lines and 150 output lines.  This will allow

   up to 150 terminals, 64 of which can be active at once. The other eight

   inputs can be other standard video signals, or gray scale or color mixed

   video produced from some group of the 64 standard channels.  



   6.  From the video switch, a coaxial cable goes to each display, which is

   an ordinary TV monitor. Possibly  some small number of good gray scale

   and color monitors are wanted.



   7.  With each monitor is a keyboard, not physically attached to the

   monitor.  The Stanford keyboards made by Microswitch (we have a sample)

   are a leading candidate, but further investigation is necessary.











							Displays      Page 4



   8.  The keyboards are connected (possibly by multiwire ribbon cable,

   possibly bit serial on coax) to a keyboard multiplexor, which in turn is

   connected to the PDP-11.

	   This completes the description of the initial components of the

   proposed system.  Here are some remarks on its operation.



   1.  At first, characters and vectors will be written into display memory

   by PDP-11 programs.  Writing a character consists of copying a bit

   pattern into display memory from the PDP-11's private memory. Because

   shifting and masking is required, a PDP-11 program might require, in the

   worst case, as much as 150 microseconds to write a character, giving the

   system as a whole the ability to write only 7000 characters per second,

   shared among 64 terminals.  This averages to about 100 characters per

   second, and will be adequate for reasonable, although not great service.

   Whether performance will be improved by increasing this depends on the

   following presently unanswered questions:

	   a.  How active will the users be?  How much do they just sit?

	   How much time do they spend doing input that requires the display

	   of just 1/3 characters per second?  How much time do they spend

	   page turning which might like a 2400 character page  twice a

	   second?  As people get more used to reading documents on the

	   system, page turning usage will increase.

	   b.  What other bottlenecks are there?  At present, the PDP-10's

	   and MULTICS massage each character outputted so many times, that

	   it is not clear these machines would put out more than 7000







							Displays      Page 5



	   characters per second even if the display system were infinitely

	   fast.  If (when) increased speed is wanted, however, a simple

	   block transfer shifter/masker can be put on the memory buss,

	   making characters at least three times as fast as direct

	   programming, and more than one could be added.



   2.  The vector generator situation is very similar.  PDP-11 programs will

   be provided for writing vectors onto the raster memory, but this will

   probably limit some applications.  A vector generator under control of

   the PDP-11 can easily be added.  Besides all this, it is possible to add

   PDP-11 or other processors to the memory buss so that a channel might

   have the full attention of a processor.



   3.  Although super-display capablility was not part of the charge for

   this committee, it turns out that 1000 line TV's can be supported in this

   system by providing additional video timing generation and making it

   possible to take four 16k blocks so as to have a 1024K raster.  The shift

   registers will be fast enough for this.  The 25 megahertz signals

   required for 1024 line TV present some electronic problems, however, so

   that experiment is required before a large scale attempt is made to pipe

   1024 line TV throughout the building.  Cheaper memories will make this

   increasingly attractive.



   4.  On the other hand, memory requirements can be reduced for character

   only displays by designing suitable on the fly character generators.







							Displays      Page 6



   This will permit more channels to be allocated to 1024 line use, but is

   not proposed as part of the initial system.  


							Displays      Page 7



   Furniture

	   Each display in an office consists of the keyboard and monitor,

   each having a power and a signal connection.  The keyboard power

   connection is a wall mounted power supply.  The ones Stanford uses cost

   $28 apiece.  We recommend also the use of metal tables with wings costing

   about $20 and a display rack sitting on the table that holds the display

   at a convenient viewing angle.  The rack cost about $15.


							Displays      Page 8







				  Procurement

   1.  The longest procurement item is the PDP-11/45 for which a (revocable)

   letter of intent has been placed.  Delivery on the 11/45 is eight months,

   but an interim smaller machine may be obtained.



   2.  The next longest, curiously enough, seems to be keyboards, with 12-14

   week delivery.



   3.  The approximate order of construction is:

	   1.  Semiconductor memory interface

	   2.  video timing generator

	   3.  teletype multiplexor

	   4.  Attachment of a few terminals

	   5.  At least one main computer interface

   At this point, host systems programs can start being debugged.

	   6.  Remaining computer interfaces

   The interface to MULTICS requires a high speed data line between building

   39 and Tech Square.  Once these are in, a 64 terminal system can be

   operational. 

	   7.  Building wiring

	   8.  Video switch

   Other construction can wait until the system is operational and we see

   where the shoe pinches.







						       Displays      Page 9



	   A related project should be to have something like the Xerox LDX

   printer on each floor so that hard copy from all computers can be

   obtained.


						       Displays      Page 10



				 Cost Estimate



   1024K 16 bit words 450 ns. memory		   146K
   (based on Yale purchase)

   150 keyboards at $205			   30.75K
   (based on quote from Microswitch)

   150 monitors at $190
   (based on quote from Ball Brothers)  	   28.5K

   PDP-11/45 with 16k memory			   26K
   (based on quote from DEC)

   Coax at 200 ft per console at 4.5 cents/ft      1.5K
   (we mention this since some people propose
   elaborate schemes to reduce cable costs)

   Video switch 				   20K
   (based on Stanford experience)

   Memory buss and video generator		   15K

   Keyboard multiplexor 			   10K

   Computer interfaces  			   25K
   (5K per PDP-10 and 10K for MULTICS)

   Furniture and power supplies 		   8K



						   ←←←←←←←
   Total not counting labor and in house design    287K
   Contingencies 20K				   20K
						   ←←←←←←←
						   307K