'1' _ HISTORY OF NSA GENERAL-PURPOSE ELECTRONIC DIGITAL COMPUTERS 1964 'lppro ed Fe F elea 6 b l r I 3A or J2-0D-2004 FIJI C I' ' # 41 02 _ HISTORY OF NSA GENERAL-PURPOSE ELECTRONIC DIGITAL COMPUTERS By Samuel S Snyder 1964 Department of Defense Washington D C 20301 -FOR OFFICIAL USE ONLY r PREFACE The author has attempted to write this material so that it will be easily understood by those who have had only limited experience with computers To aid those readers several terms and concepts have been defined and Chapter 1 includes a brief discussion of principles of computer operation programming data-preparation problems and automatic programming Engineering terminology has been held to a minimum and the history of programmer training personnel and organizational growth and-the like has not been treated To some small extent the comments on operational utility bring out the very real usefulness of computers for the solution of data-processing problems The cutoff date for eveift s-·-related -he-re---was-the end of December 1963 s s s ii TABLE OF CONTF NTS CHAPTER 1 -- BACKGROUND 'Description Page Punched Card Equipment and the Computer - - Computers in NSA - - - - - - - - Computer Principles - - - - - - - - - - - Programming Principles - - - - - - - - Data Preparation - - - - - - - - Automatic Programming - - - - - - - - - - Special-Purpose Attachments - - - - - Impact of NSA on Commercial Computer Developments CHAPTER 2 -- 2 4 4 4 e 6 AGENCY-SPONSORED COMPUTERS ATLAS I and ABEL ATLAS II - - - - - - - ABNER and BAKER - NOMAD - - - - - - - - - - - SOLO - - - - - - BOGART - - - - - CUB - - - - - - - - - - - UNIVAC l224A CRISPI - - - HARVEST - - - - - - - HARVEST Modes of Operation __ Ar i-thmetic Mode·- - - Streaming Mode - - - Indexing and Setup Streaming Instructions Adjustments - - - - - Special Memory Features TRACTOR - - - - - - - - HARVEST Operational System CHAPTER 1 2 - - - 8 - - - - - - - 13 - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - HOPS - - - - - 14 28 - - 29 - - 31 - - 36 - - 36 39 - - 46 4-6-- 55 58 S9 59 - - 59 -' - 60 - - l 3 -- COMMERCIAL COMPUTERS International Business Machines Corporation - ·6 IBM-701 - - - - - - - - - - 65 IBM-702 66 IBM-70S - - - - - - - - - 66 67 IBM-704 - - - 67 IBM-650 - - - - - IBM-1401 - - - - - - - - - - IBM-1410 IBM-7090 - - - - - - - - 71 71 76 iii TABLE OF CONTENTS CONTINUED Page Description General Precision Inc - - - LGP-30 - - - - - - Control Data Corporation - - - - - CDC-1604 - - - - - - - - - - - - CDC-l60A - - - - - - - CHAPTER 4 -- - - - 76 - - - - - - 76 - - - 78 - - - - - - 78 82 REMOTE-OPERATED COMPUTERS ROGUE ALWAC IIIE - - ROB ROY BOGART - - - - RYE UNIVAC 490 - - - - - - - - - - - - 85 88 89 APPENDIX Table 1 Chronological Listing 2 ATLAS I Instruction Code II 3 U S Electronic Computer in 1947 - - - - - - - 4 ABNER Instruction Code References - - - - - - - - - - - - - - - - - - - Activity - - - - - - - - - 93 96 - - - 97 - - - 98 - - - 99 - ---- _ -- _-- 'fLl ' USTRATx ONS Figure 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Page Description Block Diagram of Digital Computer - Typical Problem Flow Chart - - - - - ATLAS I Console - - - - - - - - ATLAS I Main Frame with Input Tape Reader in Foreground Console in Back ABEL - - - - - - - - - - - - - - ATLAS II Main Frame Partial View - ATLAS II Console - - - - - - - - - - ATLAS II Input-Output - - - - - Mercury Delay Line Diagrammatic ABNER 1 Memory Cabinet - - - ABNER 1 Console Panel - - - - - - ABNER 2 Console - - - - - - - - - _ 2 Raytheon Tape Drives ABNER 2 Main Frame - - - - - iv ¥B 4 i@ tiA - - 3 5 9 9 11 - - 15 - - 15 16 18 - - 20 - - 24 - - 25 25 26 Illustrations continued Figure 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 Page Description BAKER - - - - - - - - - - - - - SOLO - - - - - - - - - - - - BOGART Console with IBM 727 Tape Drives BOGART Serial 2 - - - - - - - - - - - - - 26 32 34 35 CUB - - - - - - - - - - - - - - - - - - - 37 Partial View of CRISPI Showing Paper-Tape Input UNIVAC 1224A and liB Console - - - 37 HARVEST System Block Diagram - - - - - - - 44 HARVEST Operator's Console Showing IBM 1403 Line Printer Right Background - - - 47 HARVEST Operating Area General View Showing TRACTOR in Left Background - - 48 HARVEST Maintenance and Engineering Consoles Arithmetic and Logic Unit left Streaming Unit right - - - - - - 49 HARVEST -- The 16 Frames of the Central processing unit - - - - - - - - - - - - - 50 HARVEST Tape Control Units and Tape Drives 51 HARVEST -- The 6 Large Memories - - - - - 52 HARVEST -- One unit of the Fast Memory - - 53 HARVEST -- TRACTOR Cartridge Handler - - - 54 HARVEST Streaming Data Paths - - - - - - - 56 IBM-70S Showing SINBAD Control Panel at Right of 705 Console - - - - - - 68 IBM-70S Core Memory - - - - - - - - - - - 69 IBM-704 - - - - - - - - - - - - 70 - - - 70 IBM-704 Core-·Memory-· ·--------· IBM-6S0 General View - - - - - - - - - - - 72 IBM-GSO Console - - - - - - - - 72 IBM-GSO Disk Storage - - - - - - - - 73 IBM-6S0 Inquiry Station - - - - - 74 IBM-140l - - - - - - - - - - 75 IBM-7090 Console - 77 LGP-30 - - - - - - - 79 WELCHER - - - - 80 CDC-1604 - - - - - - - - 81 Control Data l60-A Computer - 83 ALWAC IIIE - - - - - - - - - - 87 ROB ROY Outstation - - - - - - - - - - - - 90 ROB ROY Control Panel - - - - - - - - - - 90 UNIVAC 490 RYE General View - - - - 91 v CHAPTER 1 BACKGROUND Most writers on digital computer development tend to start with discussions of the abacus and continue with Charles Babbage's Analytic Engine the desk calculator and the big relay calculators of Bell Laboratories and Harvard Indeed the development of computing machinery did follow such a path But the role of computers at NSA can be better appreciated when considered from the viewpoint of application Punched Card Equipment and the computer The extraordinary versatility and efficiency of electronic computers have made them useful in handling almost every class of data-processing and analytic problem From this point of view and in this respect at NSA punched-card equipment -keypunch reproducer sorter collator and tabulator -- could be called the forerunners of the electronic computer For the 15 years beginning about 1935 NSA's predecessors used punched-card equipment to attack wider and wider ranges of problems During this time many special-purpose machines were also built including some designed as attachments to punched-card equipment The USe of punched card equipments as general-purpose tools continued to grow until by the end of World War II 750 machines had been installed Punched-card equipment -- keypunch reproducer sorter ---co llator and tabulator -- were the forerUni'ferS-b- the--elec­ tronic computer in every respect excepting speed and automatic operation This is true because of the use of the punched-card as a unit record flexibility of plugboard together with switching capabilities in each machine and versatility nherent in successive card passes through different equipments The following types of data analyses could be done using punched-card equipments expanding reprOducing Reproducer distributing sorting Sorter merging selecting Collator counting printing Tabulator The general-purpose computer logically corresponds to punched-card equipment in that a variety of elementary operations the order code can be combined in a variety of ways 1 ' ac the program' and applied to unit records data stored in specific memory locations The versatility is further increased by treating the program steps instructions as data themselves and eliminating the necessity for manual operations between successive logical operations Finally the great difference in speed made possible by electronic circuitry in the large computers has virtually eltminated the punched-card approach as our principal general-purpose tool computers in NSA With the earliest design work on computers in 1946 came the realization of the potential usefulness of such machines for Agency purposes Quite probably this Agency's predecessors were the first to develop sophisticated analytic applications of such machines This story is detailed later in the-discussions of ATLAS I and ABNER The use of computers by NSA has increased considerably -beginning with one of the first machines in the country installed in December 1950 NSA's computer installation probably ranks among the in the country The chronological listing Table 1 page 93 shows these acquisitions in the order of their installation and includes references to the page in the text where each computer is discussed larg st computer Principles - -- - Most modern electronic digital computers beginni ricj -with the-Sticcfessors to the famous ENIAC have been composed of four principal units input-output storage arithmetic and control fig 1 Data and instructions prepared in advance on some machine-readable medium punched cards tape and the like are fed into the machine system and the data is operated on automatically according to the steps specified in the instructions Such instructions are stored in the same internal storage medium as the problem data but are interpreted by the control unit An instruction normally indicates 1 the operation to be performed by the arithmetic unit 2 the location in storage its address of one or more operands and 3 the address of the result of the operation One of the reasons for the great flexibility of the modern computer is its ability to modify its own instructions or the course of problem solution depending upon intermediate results A variation of this capability is the technique of causing the incoming data-stream characters themselves to form addresses for insertion in skeleton instructions later to be executed r 2 __ __ __ - III lI l __ t ' 31'• •fi' ' -· IP lgl-I TII S I I@ il•• e JLl il l 1 • CONTROL # - I I ' ' ---'-- - --- INPUT- STORAGE OUTPUT ARITHMETIC Data Paths _ _ _ _ Control Patba Figure I -Block Diagram of Digital Computer 3 r 1s 1 I programming principles Any analytic or data-processing requirement however complex can be logically reduced to a succession of elementary operations that follow from binary decision-points yesor-no choices provided knowledge exists of a logical method of attack A flow chart fig 2 for a sample problem shows the use of such dichotomous or two-choice logic to break a complex problem into the major steps that will lead to possible solution In similar fashion each of such major steps is susceptible to further breakdown into smaller more detailed steps and so on down to the individual elementary machine operations that are the smallest logical unit and that correspond to the individual computer instructions Data Preparation 1r i An important and essential aspect of the use of any data processing machines is the initial step -- the preparation of data for input In connection with the general-purpose use of punched-card equipment the key-punch has been mentioned as providing eans of transforming information into a form that can be read by other machines The key-punch perforates cards in specific positions to represent data under study Punched paper tape is also used as machine input and is usually punched in 5- or 7-level codes Recently the use of magnetic tape has increased although it must usually be prepared from punched cards or··ta·pe· ···- I'Wo points should be emphasized 1 Our method for going from hard copy to machine-readable mediums has not improved -- a small army of key-punchers and paper-tape punchers is still employed 2 The existence of several media here and in the field requires the use of large quantities of equipment and time just to convert from one medium to another frequently several times within one job Many machines including several general-purpose computers have been built for the sole purpose of data conversion and rearrangement Automatic programming 1 I In the preceding'brief description of the principal parts of the typical computer and of organizing a problem for computer attack the necessity for a breakdown into elementary operations was pointed out Most of these elementary operations are more detailed -- require much finer attention -- than people are accustomed to When a workable program 4 START---- READ IN NEXT TEXT EXAMINE IDENTIFY FIRST NEXT CHARACTER ADD 11 TO MEMORY CELL FOR THAT CHARACTER ADD 1 TO MEMORY CELL FOR TOTAL CHARACTERS NO YES SET UP AND EXECUTE PRINT ROUTINE - CLEAR CHARACTER COUNT CELLS AND TOTAL CELL NO YES STOP FIGURE 2 TYPICAL PROBLEM FLOW CHART _ 'FREQUENCY DISTRIBUTION OF ALPHABETIC TEXT 5 or routine or subroutine as smaller subdivisions are called is completed it is natural to try to avoid the necessity of repeating the whole operation the next time this particular function is req ired As a result routines of standard functions are commonly stored in Jllibraries and special programs called Jlcompilers perform the function of combining several such routines and adjusting addresses for varying conditions To facilitate the writing of machine-language instructions the instruction codes of many computers use mnemonics that must later be converted into the individual computer words Programs for performing this conversion automatically are called assemblers Higher forms of languages have been developed that use a combination of rigorous English and mathematical notation to express functions and parameters of problems and thus release the analyst programmer from becoming a full-time machine specialist Programs for transforming these higher languages are called translators Jl they generate actual computer programs ready to execute a problem solution Special programs for regulating overall computer operations keeping records monitoring various types of simultaneous operations and for many other purposes are called II supervisors or executive routines In the past programmers in NSA have been less inclined to use these techniques than computer users in other fields primarily because commercially developed systems have been unsuited to the specialized nature of Agency operations More recently however such techniques have become the rule and the great ALPHA language-development effort required for HAR VEST for example will certainly have -a far-reacning- effect on future large systems Special-Purpose Attachments In large-scale equipment planning more and more use is being made of the combination of several kinds of equipment into single operating complexes Act uallY this procedure parallels what was done many years ago when attachments were built to function with the IBM Tabulator In 1954 the FARMER proposal first suggested the idea of special attachments to a computer and the design of HARVEST made specific provision for the attachment of external function devices The currently popUlar application of this principle is the designing of highly efficient counting comparing or other processing devices for attachment to a general-purpose computer for use in sophisticated attacks on complicated problems to achieve greater efficiency than would be possible using the computer alone Impact of NSA on Commercial Computer Developments In addition to contracts for computer purchase or leaSe and the development of engineering techniques the primary 6 f ' influence of NSA on industry has been felt in those instances where technical leadership or management foresight has influenced or led directly to industrial computer pioneering A few examples follow First the sponsorship of the ATLAS I by NSA's Navy predecessor led directly to placing Engineering Research Associates on a firm footing as one of the country's pioneering computer firms ERA's commercial 1101 and 1103 computers were almost identical to ATLAS I and ATLAS II respectively With the merger of ERA and the Eckert-Mauchly Corporation into the Remington-Rand-lmIVAC organization the 1101 and 1103 became a part of the UNIVAC line Through the National Bureau of Standards NSA's Army predecessor contracted for the construction of a mercury delay line memory for ABNER The Technitrol Corporation not only built memories for ABNER and the National Bureau of Standards' SEAC but later constructed the second ABNER and received a series of contracts for equipment using similar components Technitrol went on to market a varied line of special and general-use equipment much of it based on this experience Another example is NSA's NOMAD contract with Raytheon Corporation The NOMAD design was the direct forerunner of the Datama ic lOOO which was transformed into one of the early machines in the Honeywell line The realization in 1955 that the transistor had a bright future - _ as an electron·ic component led to this Agency's sponsorship of the construction by Philco Corporation of NSA's first practical transistorized computer SOLO Marketed commercially as the Transac 8-1000 it became the world's first commercial transistorized computer Beginning with the EDPM Type 701 the Agency-IBM experience is very impressive Probably very few realize that IBM based much of its early computer philosophy upon NSA's experience with large data-processing problems Certain features of the 704 representing improvements on the 701 resulted from sugges ions by NSA personnel similarly the 702 was improved to become the 705 The PLANTATION later RANCHO study contract for HARVEST design as well as NSA support for memory and magnetic tape researches not only influenced IBM's 7030 STRETCH but also resulted in solving a number of fundamental logic and processing problems never before confronted in the computer field Finally the LIGHTNING researches into ultra-high-speed computer circuits and components -- work done under contract by IBM RemingtonRand-UNIVAC RCA and others -- have undoubtedly had the greatest influence in the development of new computers and other digital equipment with basic operation times measured in nanoseconds billionths of a second 7 ··r ' ItI z ' _j£ ' P - - r m r CHAPTER 2 AGENCY-SPONSORED COMPUTERS In addition to operating an extensive computer installation NSA has been a prime sponsor of computer developments Of the nine electronic computer projects discussed in the following sections two ABNER and CUB were designed and built by NSA or predecessor personnel The rest were built under contract following logical designs conceived or inspired by NSA personnel ATLAS I and ABEL r In the summer of 1946 Moore School of Electrical Engineering University of Pennsylvania sponsored a series of lectures on electronic computers Among those who attended was LCDR James T Pendergrass a young mathematician at Navy's Communications Supplementary Activities Washington CSAW As a result of hearing the computer proposals described at that time LCDR Pendergrass prepared a report in which he proposed that the Navy acquire such a machine His report included a general description of the proposed machine's logic its code of instructions and coded examples of typical problem solutions The particular design advocated was of the' class known as one-address logic and was based on the I A S Computer then being developed at the Institute for Advanced Study Princeton New Jersey CSAW went ahead with plans for construction of ATLAS I and CSAW personnel perfected most of the basic algoritluns 6 r de'tcfir-Eicr--machine logic for executing'-' each instruction The contractor Engineering Research Associates of St Paul Minnesota used MIT's WHIRLWIND I logical design to a great extent ATLAS I was delivered on 8 December 1950 and was running before Christmas It employed parallel circuitry contained approximately 2700 tubes and cost about $950 000 Figures 3 and 4 show views of this machine Although the original proposal had specified that the ATLAS high-speed internal memory was to utilize Selectron tubes a decision was made to substitute a l6 384-word magnetic drum of a type similar to one already placed in operation by E R A in other special-purpose equipment built for the Navy Also whereas I£DR Pendergrass's proposal described a machine with 36-bit 2 words each word containing two one-address instructions the ATLAS I design was finally based upon 24-bit r 1 Selectron was the specially-designed electrostatic storage tube under development by Jan Rajchman at RCA's Princeton Laboratories The development never achieved sufficient reliability for use as a practical computer memory 2 bit -- binary digit 8 V • '5 t 6£ ft s c 4 5' ' 1i F Figure 3 -ATLAS I Console Figure 4-ATLAS I Main Frame with Input Tape Reader in Foreground Console in Background 9 words each with one single-address instruction The drum memory was equipped with a flexible feature called interlace that permitted variations in address layout to be made for each program using a plugboard setup In effect lines around the drum were renumbered by transposing certain bits of the address selection register in order to allow longer or shorter intervals between successive effective addresses according to the particular needs of the program Careful programming in conjunction with careful planning of the interlace plugging made it possible to attain extremely high speeds for given programs compared to running time of programs written without attention to interlace Thus ac'cess timel could be reduced to 32 microseconds under the best conditions compared with an average of 8 500 microseconds or a maKimum of 17 000 microseconds In May 1951 a modification the skip feature was introduced that added flexibility to programming by allowing the program address counter to advance by intervals greater than one 9 17 33 or 65 In effect the operation of the interlace feature was made partly automatic In June 1951 the installation of a dial interlace feature made selection of the interlace plugging simpler and eliminated the necessity of actually changing plugboards Other features of ATLAS lIs original design included logical multiply instruction vector addition several forms of ordinary arithmetic instructions conditional jump instructions and an elementary input-output procedure using punched paper tape In 1952 an interesting new feature the random jump instruction was added Employing a separate electronic attachment that supplied random bits of zero or one this instruction made possible the generation of streams of random characters The photoelectric tape-reading equipment- del i- ' vered with ATLAS I was found to be quite slow 140 characters per second compared with ATLAS I internal speeds nor was it under program control It was replaced by the newly developed Ferranti photoelectric tape reader that had a speed of 300 character$ 'per second Table 2 contains the ATLAS I instruction reper toire Shortly after the completion of the ATLAS I design in 1949 a decision was made to construct a relay analog of the equipment to assist in training programmers and to debug its programs at least logically before it was delivered ABEL fig 5 was designed and constructed by CSAW personnel in about four months Logically it was identical to ATLAS I but its memory drum capacity was 2 047 words instead of 16 384 and its relay circuitry made it several hundred times slower 1 access time -- time required to locate and read a word from memory into the arithmetic unit or vice-versa 10 - ' 1 f i ' ' ' · i if • ' 6 qrn w a After an initial period of perfecting operational and maintenance techniques ABEL proved to be quite reliable It was used not only for programmer training but also for generating various statistical tables Principally because of the great difference in capacity of the memories of ATLAS I and ABEL the latter could not be effectively used to debug large-scale ATLAS programs After ATLAS I began operating at CSAW ABEL was dismantled and transferred through the Office of Naval Research to the Navy Logistics Project at the George Washington University where it went back into service After operation on the Navy Logistics Project for a time it was presented to the George Washington University School of Engineering Finally in 1963 GNU presented ABEL to the Albert Einstein High School in Silver Spring Maryland Here upon advice of Agency representatives it was dismantled for-the last time Several months before the delivery of ATLAS I an order was placed with ERA for the construction of another almost identical machine it was delivered in May 1953 In 1956 both machines were modified by the addition of high-speed corel storage in the amount of 4 096 words This addition modernized the machines and eased the programmers' burden for before the addition of the high-speed magnetic-core memory the extra effort required of programmers in putting instructions and constants in those locations in-the memory best adapted to maximum use of the interlace feature was often excessive In 1957 both ATLAS I's were moved to Fort Meade and in November 1959 both machines were taken out of operation ATLAS I Serial 1 was salvaged and the components used for other purposes - -in--March 1960- --Sefrta-r- 2 was shipped to the Anti-Submarine Research Center SACLANT NATO Forces at La Spezia Italy From a maintenance point of view both machines performed extremely well Although periods of down time for modifications and additions were necessary the machines were usually operational above 90 percent in a reporting period Many useful programs were written for ATLAS I particularly for statistical and mathematical problems Although the instructions were well-balanced for statistical and analytic uses and the machines were designed and built for reliable operation ATLAS I could not be used for problems demanding the handling of large volumes of data because of the lack of a magnetic-tape facility 1 core -- also rings or toroids of magnetic ferrite material strung at the junctions of intercrossing wires Each core represents one bit of storage capacity Thus a 1 024word memory of 24-bit words would take the form of a stack of 24 matrices each matrix being a square array of cores 32 cores to a side 12 ATLAS II Several months before ATLAS I was delivered a new task was estabiished that provided for the design of a successor ATLAS II and ATLAS I differed in the following respects 1 Storage -- Whereas ATLAS I had only drum storage ATLAS II had in addition high-speed electrostatic storage for 1 024 words • 2 Word Size -- The word size on ATLAS II was 36 bits ATLAS I word size was only 24 bits 3 Instruction Logic -- ATLAS II used two-address instruction logic ATLAS I one-address 4 Order Code -- Instruction sophistication was increased on ATLAS II 5 Input-output -- ATLAS II had program-controlled input-output instructions A lack of this capability was a serious drawback for ATLAS I Two-address instruction logic was unique among computer projects In some instructions the two addresses specified the locations of two operands in other instructions one of the addresses was an operand and the other a destination location Among the more sophisticated new instructions were the repeat instruction see ABNER page 4 several modular-arithmetic instructions a scale factor instruction and an index jump instruction ATLAS II was installed in October 1953 A second machine was also constructed and it was-delivered in December 1954 The second ATLAS II differed from the first only in that it had a high-speed ferrite-core memory rather than an electrostatic tube memory Originally the plan was to equip ATLAS II with Raytheon magnetic-tape drives using a data representation of one character per frame However a change was made to a conservative scheme for data representation involving two positions per information bit resulting in three tape frames per character Subsequently this latter representation was felt to be too inefficient for most large problems With the objective of achieving tape interchangeability and speeding up magnetic-tape operations a redesign effort was undertaken ' However this modification never reached operational effective- ness Even though somewhat handicapped by lack of an effective magnetic-tape system both ATLAS II's contributed considerably to the solution of many problems Reliability was high for both machines but Serial 1 required somewhat more maintenance attention than Serial 2 because of the electrostatic tube storage ATLAS I and ATLAS II were forerunners of the first two in a commercial line of Remington-Rand computers the UNIVAC 1101 and 1103 respectively Also ATLAS II was the logical 13 model for the first transistorized computer SOLO described on page 29 ATLAS II Serial 1 was taken out of operation in February 1960 and donated to the University of Texas Serial 2 was turned over to the Second Army for disposal in May 1962 Figs 6 7 and a show three views of ATLAS II ABNER and BAKER In Dec_ember 1946 the first Pendergrass Report see ATLAS I page 8 was received at the Army Security Agency one of NSA's predecessor agencies The potential value of electronic computers in ASA applications was recognized immediately and plans were made to consider the acquisition by ASA of a machine similar to Navy's proposed ATLAS At that time 1947 there -were four main centers of electronic computer activity see Table 3 page 97 During 1948 ASA analysts made visits to these installations attended lectures at the Bureau of Standards' Digital Computer Laboratory and performed programming experiments using ATLAS UNIVAC EDVAC and RAYDAC order codes The result of these investigations was a report favoring a design using four-address logic RAYDAC or EDVAC over the one-address logic ATLAS or UNIVAC Four-address logic was favored for the following reasons 1 Time estimates for executing representative operations showed higher s p e e d s - - - - - ___ __ - - 2 A philosophic convTcfion that the bulk of computer operations in ASA applications would be characterized by relatively simple operations on many data items rather than more complex operations upon data retained in the accumulator In addition UNIVAC was rejected for Agency use because it used decimal arithmetic internally and could only manipulate data as alphanumeric characters Because of the many examples of Agency processes using binary notation a binary machine was felt to be a definite requirement Based on the foregoing report recommendation attempts were made to get estimates and commitments for building such a machine The Raytheon Corporation was already committed to complete a first machine for Navy's Bureau of Aeronautics and would not promise delivery of a second machine in less than three years Moore school had undertaken to build EDVAC for Army Ordnance and could not consider offers for a possible second machine In May 1948 Reeves Instrument Corporation announced that it was entering the digital computer field with REEVAC a proposed machine based on EDVAC design The basic EDVAC logic provided-for 45-bit words and a maximum of 16 four-- 14 7 - ' • 7- - if y ' - ' l - I - - ' - • itJ Figure 6 -ATLAS II Main Frame Partial View I ' - ' ' iiiiiii iiiiiiiiiiiiiiiiiiit- Figure 7 -ATLAS II Console A ee' P P f£ t 15 II Input-0utput gure Fi 16 1 a address instructions Its l024-word memory was to be composed of mercury delay lines with minimum access time being 48 microseconds and ac ual access varying between 48 and 384 microseconds Dr Samuel Lubkin formerly at Moore School headed the effort at Reeves His proposed design contained several improvements upon the design of the EDVAC and his price and promised date of delivery were attractive But just as a contract with the Reeves Corporation was taking form the company abandoned its plans to produce digital computers and released Dr Lubkin to the staff of the National Bureau of Standards Soon after in August 1948 the Army Security Agency 4 the NBS concluded an agreement which provided that the Bu eauwould produce a design for a computer that ASA engineers would build NBS simultaneously would be designing and building a similar computer for themselves -- one of more modest capabilities Also NBS was to place the order for the mercury delay line memories for both machines Shortly thereafter several meetings were held to settle on the functional design of the ASA machine and to orient ASA engineers in their early design efforts NBS engineers gave a series of lectures on digital computer logic The proposed ASA machine was called ABNER ABNER's mercury delay line memory was of particular interest The acoustic delay line was developed around 1946 at the Moore School by Eckert Sheppard and Sharpless during planning for EDVAC A succession of pUlses signal or no-signal travels through an acoustic medium say mercury from one end to the other of a delay line - -F2gure -9 --·shows a diagrammatic view of a mercury delay line At the input end of the line is a crystal that converts an electrical pulse to a mechanical wave which travels through the mercury to the other end where another crystal reconverts it to an electrical signal The series of electrical signals is recirculated back to input after pass'ing through detector amplifier and driver circuits to restore the shape and strength of the pulses Also in the part of the cycle external to the delay line are input and output circuits and clock pulses for synchronization In mercury the pulses travel at the speed of sound which is much slower than the speed of electrical signals and thus the delay in going from one end of the line to the other constitutes a form of storage In ABNER the mercury tank was a glass tube about two feet long the delay time was 384 microseconds or eight words of 4B bitsl at one-megacycle-per-second rate Thus the 1 024 1 The actual computer word length was 45 bits allowing a 3-microsecond interval between words 17 r • i '_- 1L ' lo 'I ' r i2 • 'al '- ' - • i' i • - I • • ••• - '- ••• ' ' _ tE£ • s _' • r ' ' WORD CYCLE POSITIONS ------- 7 6 5 4 3 2 1 0 CLOCK ------ JE ---l INPUT AMPLI FfER OUTPUT FIGURE 9 MERCURY DELAY LINE DIAGRAMMATIC r 18 words were contained in two cabinets holding 64 mercury delay lines each Figure 10 shows one side of one such cabinet containing 32 tubes of mercury In the figure the actual glass tubes are not visible being inside the metal cases shown in the upper part of the picture Because of the great variation between minimum access time and the maximum efforts were made to minimize such delays Mention has been made above of the improvements possible in ATLAS using ingenuity together with the interlace feature when special care was taken in placement of words on the drum page 10 In the case of ABNER the savings by careful programming was not as great but worth the effort Referring to the diagram of Figure 9 a word at cycle position 0 would be immediately available for readout access time 48 microseconds while a word at cycle position 7 would take 8 x 48 or 384 microseconds to be read or written into Therefore instruction words and operands would be carefully placed at memory locations that would be immediately available if possible bearing in mind cycle positions the actual memory address modulo 8 and the known execution times of instructions For the next six or eight months ASA engineers working closely with several progr ers moved at full speed to perfect the engineering logic of the original set of instructions and devised several additional refinements Also primarily to acquire experience they designed and built an input decimalto-binary conv ers i on device About this time July 1949 --·it-- -became apparent to ASA that NBS would not be able to complete the design of ABNER in a reasonable time principally because of pressure to expedite completion of their own machine SEAC So the decision was made to go ahead and build ABNER without waiting for NBS ASA engineers estimated that they could build the machine in two years Meanwhile the programmers who had made programming experiments using the several available computer designs became convinced that Agency operations justified flexibilities in data manipulation not available in these designs or in ABNERas it was conceived at the time They began work on a set of special computer instructions to be considered for inclusion in a future improved computer Three principal classes of operations required special attention character transformations data-stream manipulations and paired stream comparisons These three lines of attention became interrelated and produced a combined solution that is quite interesting and historically significant This result was possible because of the intimate working relationship between the programmers and engineers 19 o r Figure 10 -- ABNER Serial 1 Memory Cabinet 20 _·2 MMtW WAi l •• • ' r - • • ' 34 ' tiM 2 4 r • r These new features were incorporated in the machine under construction rather than waiting for a successor computer The following describes briefly how the three requirements were implemented 1 Character transformations were treated in the computer as modular addition that is the corresponding alphanumeric characters of two words are summed modulo n where the value of n varies from 2 to 32 For situations dealing with character sets containing more than 32 characters two additional instructions were devised one that dealt with two streams of lO-bit characters four in each word and another in which the pairs of characters to be acted upon were within the same word 2 Data stream manipulations originated as a requirement to transfer groups of words among different portions of internal memory The first version was thus a block transfer of a fixed number of words and was based upon a modification of the Shiftand-Extract instruction so as to repeat the instruction a fixed number of times The capability of executing such an instruction any number of times came about with the development of the Halt feature This feature originally was visualized as a means of automatically monitoring computer instruction executions so as to signal the completion of a specific routine Using the Halt instruction to signal the completion of the final execution of a repeating instruction and extending the repetition feature to apply to most ABNER instructions made these two features -- Halt and Repetition -- completely general and yielded a flexible basis for manipulating streams of data 3 Paired stream comparison was specified by the programmers as deserving special consideration because of the relative inefficiency of standard computer techniques for performing character-by-character comparison operations The engineers outdid themselves in satisfying this particular requirement and the resulting instruction was undoubtedly the most sophisticated ever proposed for a general-purpose computer Nicknamed Swish to depict its action this instruction accomplished the logical equivalent of a complete high-speed comparator In effect it a passed two streams of five-bit characters from memory through an analytic unit b compared groups of characters group size between 1 and 63 characters for coincidence c counted the number of group coincidences d stored the coincidence count in a memory location specified 21 in the same instruction and te restored one of the streams to memory at a specified offset from its original place in memory offset between zero and eight characters One significant result of experience with ABNER and the Swish instruction has been its influence on the design of the Streaming units of HARVEST When planning ABNER's input-output features the designers placed great emphasis on achieving flexibility sometimes at the expense of speed The result was an input-output capability more complete and varied than that of any computer commercially available at the time This included 1 Up to six magnetic tape drives -- These could be operated for reading or writing in both directions with words read into or out of m ry in ascending or decending order or not in consecutive order at all Block size was eight words Tapes were operable either under computer control or off-line for conversion from or to other media 2 Punched paper-tape reader and punch -- A excel electric typewriter with a tape reader and punch was first used later the tape reader was replaced bya Ferranti photoelectric reader with a much greater speed 300 characters per second 3 Punched-card reader and punch -- An IBM collator for reading punched cards and a modified card punch were operable either under computer contrQ or o£f line using the converting unit 4 converting unit -- This unit was designed to accomplish data conversion between almost any two media inclUding punched paper tape punched cards 'magnetic tape and ABNER storage It included a plugboard for performing simple character-for-character substitution upon input or output Its principal utility in practice was as an off-line converter between magnetic tape and other media 5 Input output typewriter -- Direction of insertion of characters into memory words could be chosen and ' substitution plugging on output characters was available 6 Console -- Although physically very small l2xl4xlO inches the console provided great flexibility for manually communicating with the computer and its various H 1 A Navy-sponsored development manufactured by Commercial Controls Corporation 22 registers Also a novel feature was its built-in binary-to-decimal and decimal-to-binary conversion equipment for dealing with 1l1emory addresses in ei er aeci al or binary' form An array of lights indicated the status of 1l1achine operations ana the causes of machine stops Figure 11 is a drawing of the console controls only slightly smaller than actual size Because the Agency had no prior experience in digital computer design and construction -- indeed no computers of this 1l1agnitude and speed range had yet been completed by anyone -- ASA engineers faced a number of difficult problems The mercury delay line 1l1emory was obtained from the Technitrol Company and magnetic tape drives were purchased from the Raytheon ManUfacturing Company All other components were ordered from vendors or fabricated in the Agency For example one extremely critical item electric delay lines had to be designed and fabricated by ASA engineers and technicians The power supplies console and input-output facilities were also troublesome items primarily because initially they had less reliability than most parts of the computer By September 1951 ABNER was completed and the checkThis turned out to be a very complex process partly because of inadequate instrumentation and also because of the tremendous number of variations and instruction combinations possible In April 1952 the last analytic instruction was checked out The complete list· of'--ABNER- instructions is reproduced-in- Table 4 ing out of individual features got under way In April 1955 a second ABNER was delivered constructed under contract by Technitrol Engineering Corporation Logically it was a copy of ABNER Serial 1 but it used quartz instead of mercury in the memory's delay lines An additional innovation was the inclusion of 128 words of rapid access or scat memory as well as 1024 additional words of main memory Its engineering and layout were considerably better and the design of its console was entirely different A latemodel Remington-Rand line printer was installed later Much debugging was necessary before the system became reliable 'so it had not given much operational service when in 1958 it was moved to Fort Meade In January 1960 ABNER Serial 2 was retired from active service ABNER Serial 1 was dismantled an disposed of when the move to Fort Meade took place Fiqures 12 13 and 14 show three views of the second ABNER Compared with ATLAS I neither model of ABNER had high operational reliability although they had a number of periods of good up-time The use of dynamic serial circuitry in ABNER and a large concentration of complex analytic instruc- 23 00000 00000 OECABIT LIGHTS' ®®®® @@@@ ®CD®0 0000 ®0®® ®®®0 @ 0@ B ®®®0 ®®®0 ®®®0 ZERO @ @ @ @ @@@ £ 0000 ®®®® 0000 @@@@ SHIFT 000 BETA r- ALPHA r- ®0®® @®®0 ®© i ®®@0 ®®®0 ®®®0 @@@@ @®00 ONE @0®® ' 0000 ORDER f tp IJt -- - - _ - _ ' WRITE READ 000 - OP MODE OF OPERATION - Q - s - OECABIT REGISTER FIGURE 11 ABl 1EFl 1 24 u RUN LE PANEL -- - -- f 't a__ ------_ 1St --- - ··· ··· ____ 1 -------- - j 'i d it Figu e 12 -ABNER Serial 2 Console Figure 13 '--ABNER Serial 2 Raytheon Tape Drives 25 ' '''''' t - ' t _-- - - - wi' - - --e -' -' 1If ' 'lI- ' ' Figure l4 -ABNER Serial 2 Main Frame - 26 Figure IS -BAKER tions combined to make this machine extremely difficult to maintain For example the control of temperature of the quartz delay lines in the ne mory was always extremely critical and the instruments used forneasurement and regulation were inadequate Also the input-output equipment IBM collator Remington-Rand printer electric typewriter and Raytheon magnetic tape drives were often out of operation and computation was ineffective or erroneous when inputoutput errors arose However ABNER design and construction laid the foundation for many ilnportant later developments ABNER was among the first computers in the country to operate successfully magnetic tapes simultaneously with internal computation Its analytic instructions and other unique features made it possible to perform many specialized Agency data manipulations more efficiently than certain other computers having inherently higher speed circuitry For the same reasons it was quite popular with programmers Its Swish instruction was a model for several special-purpose machines and was a forerunner of the HARVEST Streaming units Many programs made good use of the magnetic-tape capabilities as well as the analytic instructions ABNER Serial 1 cost approximately $600 000 It contained 1 500 tubes and 25 000 diodes ABNER Serial 2 cost about $750 0_00 In the fall of 1950 a proposal was made to build a slow-speed analog of ABNER based on experience in building ABEL the analog of ATLAS I Although there was some opposition to this proposal because of ABNER's complexity the group of engineers who had built ABEL was assigned the task The new analog was called BAKER and was built using electric relays and a magnetic drum memory Although ABNER was designed to use serial circuitry entirely -- in fact many of its unique analytic features were possible because of that -- BAKER engineers went ahead with the logical design and construction of BAKER using parallel circuitry and succeeded in simulating the complete ABNER order code The job took about two years and in September 1952 BAKER was tested It was equippe with a slow-speed punched tape reader for input and an electric typewriter and tape punch for output Eventually all the logic was checked out but the machine never operated reliably enough to be of much p actical assistance in ei er training programmers or debugging ABNER programs As a demonstration of a difficult engineering design job the completion of BAKER was a remarkable achievement The use of parallel relay circuitry to simulate a complex electronic serial computer however should probably not have been attempted Figure IS is a view of BAKER NOMAD Even before NSA's first electronic computer had been placed in operation 1950 Agency analysts recognized that certain priority problems would require special features in computers to manipulate large volumes of data Planners of ABNER had made provisions for the inclusion of several Raytheon magnetic tape drives and the UNIVAC system being built for the Bureau of the Census also placed emphasis on manipulations and statistical operations upon large volumes of data However it was evident that ny jobs demanded far greater capability tha could be expected of commercially available magnetic tape systems In 1949 a study contract was awarded to Engineering Research Associates that resulted in a proposal called NOMADI providing for special-purpose sorting and datamanipulation facilities The proposal to initiate a project for the development and construction of an equipment system of this nature was approved in 1950 Several companies were invited to submit proposals in response to the Purchase Description In September 1951 the Raytheon corporation was awarded a letter contract under which work was initiated A definitive contract for the NOMAD system was executed in May 1952 on a cost-plus-fixed-fee basis Although not the lowest bid Raytheon's proposal was chosen because it demonstrated what seemed to be the best understanding of the problem to be solved and its design presented a well-thought-out solution provided Raytheon's NOMAD' proposal for an internal memory having two speed ranges and arithmetic and control units of more or less conventional logic In addition a high-speed large capacity magnetic-tape system called E rimary nternal ape torage PITS was proposed to be operated in conjunction with several sections of buffer storage The PITS constituted the principal medium for the large volume manipulations called for by the purchase description and necessary for special developmental work on tapes tape drives and tape control systerns Almost at the outset Raytheon began investigations of alternative solutions to many technical questions before settling down to system design The following is a partial list 1 Memory -- The original proposal specified mercury delay lines Magnetic cores later appeared as an attractive alternative But whether to use metallic or ferrite cores was not settled for quite some time 2 Instruction Logic -- The original Purchase De- 1 After nomad a wanderer because sorted data have no fixed addresses 28 scription called for one-address instruction logic but results of an Agency-directed study indicated that three-address instructions were preferable 3 CirCUitry -- Pulse position modulation PPM gave promise of greater reliability than conventional pulse amplitude modulation and was considered for a time however after much testing it was rejected 4 Checking -- Raytheon engineers made elaborate plans for including checking and error-correcting circuitry However Agency man gement directed a minimization of such checking features because of the high cost of the large number of additional components 5 PITS -- This large-volume magnetic·' tape system posed a number of problems -- tape handling head design tape manufacture control logic buffering and the like There was insufficient assurance that all these problems could eventually be solved In addition to technical problems serious'personnel and security dilemmas arose to plague the project As early as February 1953 one of the company's key experts on tape systems resigned later the project manager himself failed to obtain full clearance which made it difficult to explain problem background for certain of the system features Finally an evaluation of the contractor's effort in 1954 concluded with a recommendation that the contract be terminated because of prohibitive time and cost overruns as well as the likelihood that NOMAD would be obsolesc ent- at time of delivery When the books were closed the contract was shown to have cost the government approximately $3 250 000 Allowing the record to indicate that this expensive failure was due to contractor shortcomings might be more comforting but an honest assessment of the NOMAD experience indicates that Agency actions or weaknesses may have been partly to blame Among these are the following cited as possible object lessons 1 The construction contract was awarded before enough attention was devoted to design optimization 2 Insufficient investigation was made of potential contractors' abilities to perform 3 The Agency technical staff was too small to properly supervise the contract SOLO By January 1955 the substitution of the transistor for the vacuum tube in switching and power applications became so 29 widespread that the Agency created a new task PEANUT devoted to accumulating c1rcuit info%mation and to forming the basis for training other engineers- in transistor technology Later that same montb - a proposal was -made that the Agency sponsor the construction of SOLO probably the country's first transistorized large-scale electronic digital computer -- a desksized logical copy of ATLAS II The proposal receiveq the full support of the Director who directed that the goal of this development was to be the construction of forty machines for decentralized operation in analytic areas The contract to design and construct SOLO was awarded in June 1955 to the Philco Corporation then the only producer of high-speed transistors A primary objective was to demonstrate the feasibility and reliability of transistors in computer circuits in a large computer system using direct-coupled transistor logic Later subcontracts for the construction of the core memory and the power supplies were awarded to Remington-Rand-UNIVAC and Magnetic Controls Corporation respectively Although SOLO was planned as a logical copy it did differ from ATLAS II Whereas ATLAS II had a drum memory of 16 384 words in addition to its high-speed memory SOLO's internal memory was confined to 4 096 words of magnetic core storage Also since the smaller memory required only 12 bits instead of 15 to address any word the three extra bits were used as address modifiers to designate -l2 -bit po-r--tions·of an-oper-and--word This feature added flexibility and increased capacity for certain types of operations In addition SOLO had a new instruction Expand Transfer that facilitated the expansion of data packed nil characters per word into a series of lin words one character each Also input to SOLO was by a Teletype high-speed punched paper-tape-reader operating at 100 characters per second output was by a Teletype paper-tape punch or a Flexowriter at· 10 characters per second The contractor's construction effort was slowed by a number of difficulties which could be attributed partly to the unsatisfactory power supplies initially delivered by one of the subcontractors and partly to the instability of certain components of the other subcontractor's memory After several time and cost overruns the system was delivered in March 1958 it cost the government $1 007 700 Principally because of the technical shortcomings just mentioned NSA engineers spent approximately a year debugging the equipment Power supplies were replaced and the tape punch and Flexowriter were repaired 30 SOLO was composed of a com uter console and an inputoutput console fig 16 and contained 8 000 transistors 3 7aO resistors 200 capacitors and 8 vacuum tubes In spite cf the diffioulties SOLO finally operated very reliably but original ideas about its use were modified Instead of many oopies being made for operational use only one SOLO was built and its use ltmited to research For example it was used as the vehicle' for testing and modifying the RCA agnetic tape drives under development for RID These three RCA magnetic tape transports had special features to provide flexibi1ities not available in commercial systems Of course this had the effect of extending SOLO storage capacity so that large experimental programs could be run Also a Potter tape transport was added to increase SOLO's inputoutput speed Making these modifications locally attendant supervision repairing and debugging occupied some 35 000 n-hourg spread over several years Although much effort and time were devoted to the project to attach and test the RCA tape drives they never became operational because of difficulties and inconveniences in converting from primary input media However the Potter drive was quite effective because its tapes were interchangeable with those made on IBM Type 727 tape drives Operationally SOLO programs were written for two purposes 1 to produce empirical statistics in many problem situations -preceding--speci-al-purpose equipment design ' and 2 to simulate special-purpose equipment during design and construction SOLO was the first computer extensively used by Agency engineers to assist in engineering logic-check or design optimization The original objective of SOLO -- to prove that a reliable operational computer could be constructed using directcoupled transistor logic -- was attained In August 1963 SOLO was removed from regular oper tion and turned over to a machine-processing group for t aining purposes BOGART In Chapter 1 page 4 the problems attending the preparation of input data for computer treatment have been touched on The difficulties inherent in Agency activities --lack of control of source material variations in communications practices rigid formatting requirementF of computer programs -had resulted in the construction of many different types of conversion equipment In 1952 and 1953 suggestions were made for using especially designed digital computers for data conversion and editing and to clean up raw data for input to 31 1s2 fr 1 Ti I ll 11nurarwlolvhWNWul- vd- eqomama 23E Juanlarger computers During the NOMAD devel'opment for example one proposal for th e solution of its anticipated 1tlassive data preparation conversion and formatting requirements was to build an additional computer for these purposes In December 1953 a proposal for the design and construction of such an editing computer was approved The purchase description of BOGARTl specified-a- logical design based upon 7-bit words and provided for magnetic core and magnetic drum storage punched paper-tape input punched paper-tape and card output and a flexible set of 3-word instructions In July 1954 Engineering Rese rch Associates Division of RemingtonRand was awarded a contract to build two of these machines The decision was made to use diode and magnetic core logic for arithmetic and control circuitry the basic memory cycle time was to be 20 micros·econds Several months of programming experimentation showed that the small 7-bit word size and the awkwardness of handlinga 3-word group for every instruction caused much difficulty for programmers and put a strain on the memory capacity In July 1955 a modification was made that provided for several index registers and larger word size 24 bits with flexibility in dealing with 9-n·it portions of any--wor-d • --Now· also -four machines instead of two were to be constructed and equipped with IBM Type 727 magnetic-tape drives Serial 1 was delivered in July 1957 Serials 2 and 3 in November 1957 and Serial 4 in January 1958 SUbsequently the pilot model was utilized as the central computer for the remote-operated system ROB ROY so that five BOGART machines in all were delivered to the Agency BOGART has operated extremely reliably It has been in great demand for many jobs other than purely editing and data conversion In 1957 an assembly program SLAVE was created and in 1959 BARN OWL an executive routine was introduced Later BARN OWL was refined and expanded to include some compiling functions As an operating system it became known as PHOENIX The BBC Compiler based on the ABC Compiler for the IBM 704 is also available Figure 17 is a view of the BOGART console and figure 18 is a view of BOGART Serial 2 1 After John B Bogart famous city editor of the New York Sun 33 Figure 'iorxsole with IBM 1 27 Tit pe Hives 1 '_ l ' I f - 1 o •• - Figu e 18 -- BOGART Serial 2 CUB In March 1960 in connection with an R D task for the development of 2 5-megacycle miniature building-block circuit cards for electronic equipment construction the decision was made to design and construct a small test vehicle Shortly thereafter the decision was changed to a full-scale digital computer for this purpose and construction of CUB figure 19 began in June 1960 An engineering and design Jab within RID designed and constructed the entire machine and it was completed in January 1961 The CUB computer was a small general-purpose machine of limited processing capability It contained approximately 1 000 printed circuit cards and was operable at normal room temperatures using blowers for air circulation The CUB word-size was only 12 bits and each of its 32 instructions consisted of two words Its memory capacity was 4 096 words its cycle time 8 microseconds The typical flexible oneaddress repertoire of instructions -included a Repeat instruction similar to that in ATLAS II CUB was originally equipped with a photoelectric papertape reader 200 characters per second and a Flexowriter paper-tape read punch typewriter 10 characters per second Later to test the use of cards in control units for peripheral equipments an Anelex high-speed printer and two IBM Type 7330 magnetic-tape dr i ves were _ added -- These-machines wer-ethen made available for use with other Agency analytic equipment CUB was used to establish system design parameters and to assist logic designers of other machines in processing logic equations Despite its limited purpose and short construction time CUB performed quite reliably and was very useful With ABNER it was one of the two general-purpose electronic digital computers totally constructed by Agency personnel UNIVAC 1224A CRISPI UNIVAC 1224A is a name applied by the Sperry Rand Corporation to the computer that the company built for CRISPI a bauded signal receiving and processing system b inq devAloped by an RID enqineering and design group The UNIVAC l224A is the heart of CRISPI and its organization and instruction repertoire are directed toward the concurrent manipulation of 16 or more independent data-streams 36 Figure 19 --CUB s- -- - - i I J -- - 0 - _ • - Figure 20 -Partial View of CRISPI Showing Paper-Tape Input UNIVAC 1224A and B Console 37 eatures descriptive of the UNIVAC l224A are Word length 24 bits plus 2 parity bits Memory' size 16 384 words Memory o90le time 4 microseconds Number of instruotions 28 single-address Number of index registers 31 one live others in memory Input up to 16 independent data-streams output variety of outputs hereafter described Number of circuit cards approximately 1000 Physical dimensions 24 wide 27 deep 72 high Power requirement approximately 1000 watts Specifications designed to meet MIL-E-16400 Mean time between failures 1 200 hours estimated -- The logic of the input output portion of the UNIVAC 1224A provides for communication with several external devices To prevent simultaneous access to memory by more than one external device a priority system regulates accesses by prorating them fairly Present external devices include papertape input output device printing facility magnetic-tape unit on-line communication line and one to four operator consoles The paper-tape input output device provides a minimal capability and consists of a 300-frames-per-second photoelectric reader and a llO-frames-per-second paper-tape punch The printin g_f C 9ilityuses a one-way-only flow of'data and can be any device that will match the computer's electrical and logic interfaces The device presently used is a printer control unit designed to distribute up to 64 output streams to paper-tape punches on one-at-a-time page prin ers A bidirectional interface on the magnetic-tape facility permits data to be transferred from computer to tape or from tape to computer but not concurrently The magnetic-tape control unit is designed to handle up to four transports The tape format is compatible with IBM Type'729 low-density tapes The on-line input output facility permits bidirectional data exchange between two computers or between one computer and some other asynchronous device such as a communication center The operator console is the critical on-line interface equipment between the signal tuning equipment receivers demodulators and the like and the computer The console allows the operator to control the flow of data to the computer and to monitor the activity in the computer The logic interface between the console and the computer is such that a variety of operator consoles may De used To date four different consoles have been conceived two of which have been constructed 38 Two UNIVAC l224 computera have been put into use the first in June 1963 and the aecond in July 1963 Pe r or mance has- oeen excellent Four -Ulore are under construction for delivery early in 1964 Procurement of 10 -more for othez _purposes is being initiated Figure 20 a partial view of CRISPI shows the punched paper-tape input unit the UNIVAC l224A computer and the operator console type B HARVEST -- The most important characteristic of the stored-program general-purpose electronic digital computers has been their spectacular versatility -- their general usability for statistical jobs logical jobs analytic jobs and the like However because of the elementary nature of East individual computer operations this characteristic is often a disadvantage for performing certain specialized operations In other words a special-purpose machine can be built with analytic power superior to any general-purpose system assuming equal states of the engineering art On the other hand this superior power has not come easily Special equipments have well-known disadvantages -- initial delay for specification design and construction possible disappearance of problem requirement and high cost per machine Special efforts were made to reduce these disadvantages to cut down costs and delivery time In April 1954 a possible solution of this machine designer's dilemma was proposed in FARMER a system of g er pose and special-purpose equipm n __i h t would include suitable prov-isions for switching ' ' c-ontrol and data transmission FARMER would attempt to combine the advantages of both types of equip-ments minimizing or eliminating the disadvantages of both by providing 1 a powerful general-purpose computer capable of controlling the operation of special-purpose attachments as well as the flow of data through the system and 2 specialpurpose high-speed attachments built and connected as problem requirements dictated Thus ntorage input-output and statistical requirements of special problP-ms could be standardized and provided in the general-purpose portion leaving only the heart of a proposed special analytic attack to be designed and constructed Just about the time that FARMER was being considered the Agency's efforts to get the large-scale computer system NOMAD were reaching a crisis see page 28 29 The contract with Raytheon Manufacturing Company was terminated in June 1954 During that same onth an Agency committee assigned to set priori ies in equipment planning took up consideration of the FARMER proposal With current NOMAD difficulties an important factor in its discussions this committee created an ad hoc study group for the purpose of surveying the Agency's 39 needs for large-scale analytic equipment and preparing recommendations for design studies The report of this group published in November 1954 approved the ARMER proposals and emphasized the following particularly desirable characteristics for the proposed syste and its parts 1 Compatibility and uniformity 2 An increase in speed as far as practicable 3 Break-up into separate units that can be freely interconnected 4 Multiple copies of those units that are required most frequently 5 Ability to put early units into operation before other FARMER units are completed 6 Ability to incorporate new unit-types as their need appears During the next few months some effort was made to begir launching a FARMER System Detailed investigations were conducted to determine what special attachments should be considered and some high-speed circuitry research was undertaken but little of a concrete nature really got under way In May and again in-August 1955 IBM representatives contacted the Agency and described their STRETCH program of engineering and logic researches -- researches directed towards development of an advanced computing--system-hav-ing one hundred to two hundred times the power of the best existing machine These prospective features were based on the following specific developments 1 Transistor fabrication and circuitry advances in the 10-megacycle-per-second range 2 Advanced core-memory techniques showing promise of reaching access times of about one-half microsecond for small memories about 1 000 words and two microseconds for larger memories approx 16 000 words 3 Improvements in magnetic-tape handling systems making the attainment of an information transfer rate of upwards of 1 200 000 bits per second appear practicable 4 Revolutionary logical design and organization improvements -- program look-ahead automatic indexing automatic table lookup and the like - IBM representatives explained that the company had invested many millions in the research program but was now too short of funds to build a hardware system to test the logic circuitry and components Because of NSA's need for such a 40 large-scale system and because of this Agency's ability to t es-t---operate the system IBM proposed that the first STRETCH system be built 'and delivered for a fixed price of $3 500 000 Although Agency technical personnel considered this a great bargain it was rejected because our engineers were not convinced that the high-speed memory techniques were developed and proven ready to incorporate into a hardware system and also because Agency requirements for a system of this magnitude justified a special effort to insure that its design was more Agency-oriented The strong recommendation of the Agency engineers and analysts involved was that these developments should somehow be incorporated into the planned FARMER system For this reason after Agency discussion and negotiation with IBM two lS-month study contracts SILO and PLANTATION were awarded to IBM -- Under the RID study task SILO IBM's high-speed memory development was supported to the extent of $821 203 This task was supervised by a nucleus of engineers conducting advanced component researches These memory developments centered around two approaches that corresponded t o the halfmicrosecond and two-microsecond access times -- referred to at IBM as the fast and large memories respectively The so-called fast memory utilized multi-hole cores At the time this unique development was considered one of the weakest key components of the proposed system and therefore was the principal basis for initiation of SILO PLANTATION later renamed RANCHO cost $828 340 and referred to the design studies that led to the Agency-oriented computer system HARVEST During the first few months of 1956 the activit ies under this contract included physical security arrangements clearances of personnel at IBM and the creation of an Agency FARMER-HARVEST task force to indoctrinate the cleared IBM personnel in NSA problems and machine needs and to exercise technical supervision Primarily the task force described our analytic machine needs and background and wrote descriptions of typical problems tha the n9W system would be expected to solve ' After a year work on the high-performance tape system supported by IBM's limited research budget alone did not seem to be progressing as well as other hardware developments of the system Because IBM had no plan to perfect a commercial line of such tape systems and because NSA perhaps more than any other computer user urgently required such a high-performance tape system another 18-month study contract TRACTOR was awarded to expedite this development This contract began in 41 I' January 19 57 and coS± $337 443 After a period of experimentatLon with different parameters for such a system the design goal was set for an automatic cartridge-loading system having approximately 100 ti es the speed of the IBM Type 727 tape drives then in use On May 1 1957 IBM delivered the Preliminary HARVEST Manual of Operation containing proposed logical design for the HARVEST System Certain aspects of the situation confronting IBM had a definite influence on the design of HARVEST Some time before the 1955 discussions with NSA IBM had submitted proposals more or less in direct competition with Remington-RandUNIVAC to the Atomic Energy Commission's Livermore California facility fo the construction of an advanced computer system The R-R-U design was selected by AEC and the computer system LARC Livermore Automatic Research Calculator was already under way by 1955 IBM decTded to push its advanced computer researches still more energetically and its program STRETCH went ahead with several millions of company research money After NSA's negative first reaction to IBM's offer to build a computer at a fixed price ABC's Los Alamos Scientific Laboratory accepted substantially the same proposition In addition the desirability of guiding the computer design 'effort so as to result in a commercially-marketable system was of long-range interest to IBM's top management The subsequent actions to initiate design and memory studies for NSA put IBM -in the difficul't' technical position of being confronted with three possible conflicting courses of action within the same computer research activity also their staff of qualified technical people was limited The'different emphases in the three courses were briefly 1 Atomic Energy Commission required extremely high calculation speeds and the ability to solve very large systems of equations and deal with extremely large numbers 2 Emphasis in NSA was on very large volumes of input data to be processe4 rather than high-speed calculations the ability to perform complex logical operations on two streams of data at high speeds and extreme versatility in attacking Agency problems of great variety 3 Computers for the commercial market required somewhat less advanced capabilities in calculating speed and volume of data to be processed but greater emphasis on economy and reliability To arrive at a satisfactory solution to this trilemma IBM created a committee of senior computer design experts 42 - 3 1 Committee The recommendations of the resulted in a plan for tfie des1gn of the following thr-ee principal processors dubbed the camm ttee 1 A Central Processing Unit providing a special arithmetic unit together with advanced flexible control features' like efficient indexing look-ahead and interrupt capability 2 A high-speed floating point l parallel arithmetic unit primarily to serve AEC needs 3 A Streaming Processor NSA's HARVEST addition for high-speed manipulation of two input streams of characters and the returning of a resultant stream to storage Within this concept a practical computer configuration of a minimum number of units would consist of a central processing unit associated with one or more blocks of memory and a basic exchange for control of standard inputoutput equipment all connected through a bus system The addition of the floating-point unit would satisfy ABC requirements and the addition of the Streaming Processor would satisfy NSA's basic requirements To-take care of NSA's large-volume problems a high-speed exchange was proposed through which high-performance tapes TRACTOR could be connected in the system as well as optionally a high-speed disk memory The hus system_ was to be designed to permit later connec·tion of spec i al--purpose attachments if requi red An obvious advantage t o both A EC and NSA of such a philosophy of system des tgn would be the sta ndardization and qual ity contxcd of circuits and components being read for the com·r H rc 1 a market and the e l- 'T 8cted economy and availabili ty of SI arE pa rts As the design work progressed and the interaction of th o ditfenmt par'ts of the syst €' ffi was better understood eco nomy and expediency su'gg8$zted that certain 11l ' 'CWEST registers c se portions of ha rdwaJ e in the basic and that the B2- RVES' ' Systew inc 1 ude treE AEC f1 oa tinq-point at tacm-nent 'rhe simplified block diagram of Figure 21 shows these overlaps among the three principal parts of the system configu- ration 1 L Floating-point ari throetic a method of calculation in r hie h t h em a c rd 11 e a 1 t 0 ill a tic all y p 1 ace s t 11 e dec i' J1 a l O T binary point 43 -- INPUT OUTPUT EXCHANGE -- lAT ' MtlrlDry 6 nlol 16 3 4 warda ech 2 18 microeec ace time I CENTRAL I'ROCESSING FLOATING POINT ATTACHMENT HARVES'1' A'I 1'ACHMENT UNIT I I I I I I _ J I- To Input·Output Unital I r-- L-- l I Special At n 1 I I - J HIGH SPI ED EXCHANGE After the submission of the HAEYES'l' desi9n proposal in l1a y 19 57 an ad hoc group to eyaluate HARYEST was created consisting of 14 pro amme rs and tfireeengineers drawn from research and operational parts of the Aqency This group studied the design proposal and de rough time estimates for the execution by HARVEST of 15 problems considered representative of Agency needs in this category of equipment In most of these a comparable estimate for doing the job by another computer or special-purpose machine was also prepared The report of the group Qated 31 July 1957 indicated that HARVEST would be faster than the equipment currently used and in most cases by factors varying netween roo and 200 Where sorting predominated the estimated factor of speed increase over the IBM Type 705 ranged from 50 to 100 Members of the ad hoc group also made several suggestions for design improvements A later evaluation dated 3 January 1958 included comments on the reasonableness of the IBM cost quotations and a discussion of estimated efficacy of alternative equipment combinations It also recommended initiation of both contractual and local efforts towards the creation of a sophisticated programming system In April 1958 the formal purchase description for HARVEST was presented to· IBM The RANCHO study task had been extended until June 1958 and the provisions of the purchase description were worked out by personnel of NSA and IBM This purchase description included several improvements over the original HARVEST-design proposal Two letter- contracts --·- based upon IBM cost estimates and later converted to definitive contracts were drawn up for the design and construction of the HARVEST System they were ratified by IBM on 30 April 1958 A number of revisions of the original proposal were made during the design and construction of HARVEST A full account of the succession of changes was made as a part of the Contractor's final report Comment on several key features of the system as it was delivered in January 1962 is sufficient here The original high-speed 0 5 microsecond access time memory development was based upon the use of a special 3-hole core as the basic storage element This project was abandoned in May 1960 because it could not be completed in time for inclusion in the finished system In place of the 3-hole core memory another memory scheme known as 2-D was substituted it used two toroid cores per bit Completed and installed in the system access time in the fast memory is about 0 9 microsecond Still greater speed is possible by 45 overlapping memory references between the two 1 Q24-word banks so that an effective 'access time of approximately 0 5 microsecond is obtained The large memory access time of 2 18 microseconds is substantially at its original 2-microsecond goal Orders were placed for four additional 16 384word banks so that a total of six times 16 384 or 98 304 words of large memory were installed As in the case of the high-speed memory overlapping memory references among the six banks of large memory can make the effective access time substantially shorter The character streaming time in the stream processing unit originally set at 0 2 microsecond finally became 0 3 microsecond In the basic computer the time necessary for typical arithmetic operations is also longer than was originally planned The high-performance tape system TRACTOR has' met all its original goals In addition the overall TRACTOR storage capacity has been doubled by increasing the number of cartridges available at each automatic cartridge handling unit from 80 to 160 The HARVEST System delivery date of January 1962 was about 15 months later than the original target date and the cost of the system exceeded the original estimate by about 32 percent Because of HARVEST'S unique position in NSA history a semi-technical description of its principal features has been included here Obviously a full technical treatment is beyond the scope of this effort however an attempt is-maaeto gi a pictur of-HARVEST _ that shows its similarities to and differences from conventional computers Figures 22 through 29 are views of the principal portions of HARVEST HARVEST MODES OF OPERATION Primarily because of the actual hardware overlap among portions of the Central Processing unit CPU and the HARVEST attachment HPU the system operation is said to be in ither the ithrnetic mode or the Streaming m de Arithmetic Mode - In this mode the system operates similarly to conventional computers -- that is generally instructions indicate operations to be performed on data specified in terms of 64-bit words with addresses In HARVEST however the addresses can indicate not only any word in core storage but any desired number of bits beginning at any specified bit position that is a serial arithmetic unit is provided which operates on fields of data of varyinq length The kinds of arithmetic operations possible include integer operations performed in 46 Fj ure 22 -·HARVEST Operutor's C me Printer RightO O lkO' ShowinG filM J 103 ac ground 'j 'f c I I Figure 23 -- HARVEST Operating Area General View Showing TRACTOR in Left Background - iiii· iiTj1f •••IS 2 _ I••r-_ _ _ ' #' f· j '1' -- Figure 24 -HARVEST Maintenance and Engineering Consoles 1 Arithmetic and Logic Unit left 2 Streaming Unit right 49 Figure 25 HARVEST The 16 Framuu pi Us Processing UniL IQ Figure 26 wv HARVEST Tape and Tape Drives Control Units 4 Figure 27 j'v r- 1% HARVEST u The 6 Large Memorie a we Figure 28 HARVEST One Unit of the Fast Memory en - 1 J • 1lL Fi'lur8 29 _ llAlNBS' ' '1'I AC'J'0H ' l II id'J 1i I either binary or decimal representation radix conversion operations decimal-to inarY'or binary-to-decimal conversion connective operations paired bit combinations and parallel floating-point arithmetic operations The operation of the Instruction Unit I-Unit and the look-ahead Unit LA-Unit represents the most important logical advance of P EST's arithmetic mode over conventional computers This advance exists primarily because of the virtual elimination of waiting time caused by imbalance between fetching storing indexing and execution times The I-Unit fetches and indexes instructions loads the LA-Unit and monitors storage addresses interrupt operation time clock operation and operator console control The LA-Unit makes possible the overlapping of fetching of data from core storage with instruction fetching modification and execution An important part of this process is the insuring by the LA-Unit that errors in data transmission and instruction executions are caught and corrected and that the state of interrupted operations is monitored and restored if necessary Streaming Mode The HARVEST Processing Unit HPU has been characterized as the initial implementation of the FARMER principle That is it can be considered to be a special-purpose attachment operating under control of a general-purpose computer and sharing the same input-output and storage- ·facil-ities - -It violates the basic FARMER principle however in that portions of it are built as integral parts of the basic computer and therefore are not detachable On page 22 mention was made of the influence of ABNER's swish instruction on the HARVEST stream unit design In both these machines the requirements were similar to those of a file-maintenance operation A file of data must be updated from time to time by deleting incorrect information inserting or adding new data and making calculations or transformations In both ABNER and HARVEST the innovation was the provision for a simultaneous accessing of two streams of data from storage performing selected operations and delivering to storage a single result stream of data Naturally HARVEST's HPU is considerably more sophisticated logically than ABNER'S swish as well as using basically higher speed circuitry The follow g paragraphs briefly discuss some of the ways in which the HPU operates on data-streams The HPU uses two source units P and 0 that correspond to the two input streams of characters and a sink unit R that corresponds to a result stream Figure 30 is a simpli- ll 55 BUS from Memory t I p Indexing Unit 64 64 Register P f--'o -----Swil ch Matrix I Match Stalion 1 Mask 1 BUS from Memory 64 t Q Indexing Unil 64 ------Switch Matrix H Mae I 1 R isl r Q I-- I Match Slalion I f-- Indexing Table Addres A lnblcr - Tobl CowuenJ xistence slurcd in LOGiC UNIT 1 q- I I H Indexing Unil Switch Matrix - ------- Mask Match Sl lion 1 I I I r Stalislical Counter BUS lo Memory Fi ure 30 ·-HARVEST Tahle Extract Unil I H 1 I 64 r-- Indexing Btatist ical Accumulator Regisler R 64 I Streaming Data Paths 'n hold M mury -1 Match SlllLinn fied representation of the data-paths of the HPU and shows the principal units and their relati onships P and Q fetch 64-bit data-words from core storage and process the data through a switch matrix tfiat selects and regulates the release of individual bytes 1 for subsequent processing in other parts of the HPU Byte masks EM at outputs of P and Q and input to R permit ordinary bit masking Match stations MS act as monitors to detect occurrences of specified preset characters or errors during stream processing The four match stations -- at outputs of P Q the Logic Unit LU and the Table Extract Unit -- are capable also of inserting a substitute or corrected byte into the stream without delaying the stream process The Logic Unit combines two 8-bit input bytes in any of 32 possible combinations The LU can also generate several one-bit stimulus signals to be supplied to several possible strategic positions in the HPU and used for control interrupt and status indication Modular and other arithmetic transformation operations that generate an output stream of characters are performed in the LU The statistical counter SCTR is used to count any of a number of stimuli generated in various units in the EPU that have been set to react to specific conditions Associated with the counter is a register that stores a threshold Because an output of SCTR can be entered directly into result-stream R events or reactions tq eYE nts occurr ing lu ' ng streaming can affect subsequent cycles of this stream The statistical accumulator SACC receives a-bit or 16-bit fields from the Logic Unit or the Table Extract unit Both algebraic and absolute value accumulation are available and stepping-by-one can be done under control of setup or adjustment Also a threshold can be set with generation of a one-bit stimulus when the threshold is reached or exceeded by the total in SACCo Before describing the Table Reference Unit a brie f explanation of table look-up as implemented by conventional computer techniques may be helpful The technique of table look-up is one of the most interesting powerful and versatile tricks widely used for years by programmers on conventional computers Lists of employees names or numbers telephone directories ordered collections of various sorts are a few examples of tables or files which must be consulted and operated upon Because gaps in such alphabetically or numerically-ordered data are inevitable and unpre0ictahl a I byte -- a group of 1 to bits 57 straightforward process for finding individual entries is apt to be proh J bitively time-consuming The problem is further complicated because typically each job involves use of an unknown or uncontrolled stream of data to be examined The principle underlying the programming solution to such situations involves a system for utilizing the individual data items themselves to create the address in memory where the applicable extraction or insertion must be made Typically the combination of a data item with some constant would form the address which is inserted into a skeleton instruction Such a machine-made instruction would then be ready to be called into Control for execution The whole process more or less as described here using conventional instructions is far more efficient than a brute force approach which examines memory contents one at a time - In the Table Reference Unit this whole process is performed automatically under program setup control A Table Address Assembler TAA provides a means of adding bytes from either the P-stream or the Q-stream to a designated 26-bit table base address The resulting sum may be used in instructing storage to perform look-up count l or existence l functions The other i portant use of an address formed in this manner is to extract information from words in core storage using the TEU and send the information to the match unit and logic unit and if desired to the R sink unit and the statistical accumulator SACC other useful functions-are served by TEU for example in connection with indexinq Indexing and Setup -- The 'what' rather than the 'how' of arithmetic and streaming modes have been briefly discussed A feature of both modes of importance is indexing and especially in HPU the setup words In the arithmetic mode the Instruction Unit performs indexing as in most modern conventional computers that is addresses are modified prior to their execution The programmer may specify this as an addition of the value of the index word to the operand address normal or by use of the index value itself as the effective address progressive In the streaming mode however indexing is much more complex By means of parameters placed in specified fields of a series of 20 setup words the determination and regulation of sizes of bytes the extent of oVArlap if any among successive bytes the nature and length of cycles and the definition of successive levels are all made auto- 1 See Special Mamor Features page 59 58 matically The index control units generate three basi c types of patterns for parameter manipulation -- sequential nested and triangular In the sequential type each level of indexing is in control only once and is a reading level i e at a given level indexing controls the read ng or storing of bytes In the nested type each level has control many times but only the first level is a reading level The triangular type on each succeeding iteration of the level decreases or increases by one the number of iterations that the particular level performs Streaming Instructions -- As mentioned HPU instructions are preceded by setup words The principal streaming instruction that governs byte-by-byte operation is the Stream Byte-by-Byte instruction SBBB i this instruction itself contains additional parameters Most of these parameters refer to gates not shown in Figure 30 that govern data paths among the elements of the HPU and the bus In addition to the byte-by-byte streaming operations HARVEST provides the following six hybrid instructions so called because they are not byte-by-byte yet apply to the HARVEST concept of streaming SMER SSER SILS SSEL STIR SQNL Stream Merge Stream Search Stream Indirect Load or Store Stream Select Stream Take-Insert-Replace Sequential Tabl-e-·· Lookup Adjustments -- An adjustment unit monitors non-arithmetic op- erations and upon the detection of certain stimuli interrupts the normal operation and initiates special operations These special operations modify the basic indexing pattern or temporarily vary the basic process Adjustments may be used by the programmer to perform operations that cannot be performed entirely by the control circuitry of an ·individual unit Adjustments can be used to terminate setup instructions but usually normal non-arithmetic operations are resumed as soon as the special operations are completed Special Memory Features -- Among the features originally speci- fied during design studies were capabilities in core storage that would facilitate making frequency distributions with a minimum of proqramming and housekeeping steps These requirements were fulfilled by the three' features known as clearing counting and existence 1 The clear feature is useful even without the other two It permits large areas of core storage to be set ·to zero rapidly In the fast 0 9-micro59 second memory either 32 or 256 consecutive addresses can be cleared to zero with one instruction The large 2-microsecond memory can be cleared in blocks of 128 or 512 consecutive words with a single instruction 2 The count feature available only in the fast memory allows a 1 to be added to a given bit-position Several variations of count instructions permit the accumulation of numbers with a choice of maximum sizes of 8 16 or 24 bits The count feature used in conjunction with TAA and TEU is a powerful and rapid facility for preparing frequency distributions 3 The existence feature permits the mapping of large sections of memory so as to represent the fact of existence by the state of individual bit positions Thus much greater capacity is available to record the status of large data files for example if only a 'yes' or 'no' fact is required for each group In an existence operation available in both classes of core storage a 1 is stored at the selected bit-position regardless of the present contents of that position TRACTOR When it was decided to initiate the TRACTOR study contract in January 1957 a primary consideration was to bring the effective·- speedup- over current magnetic-tape -systems closeto the speed increase of the rest of the system that is TRACTOR tapes would have about 100 times the speed of IBM Type 727 magnetic tapes An additional important requirement was a'need for the capability of handling massive volumes of data in problem situations not amenable to attack by any other means The researches at IBM had pointed the way to such a capability but additional emphasis was necessary to bring these tape studies up to the same level as other parts of the proposed system By May 1957 the date of submission by IBM of the original HARVEST design proposal the broad outlines of the TRACTOR System and certain of its specifications were set Even though the contractor encountered numerous difficulties during· the following two or three years the TRACTOR portion of HARVEST finally met or exceeded most of the original detailed specifications A description of the principal characteristics follows There are three automatic cartridge handling units each of which has facilities for automatically seeking and extracting a specified cartridge from its position in two storage units under program control Each cartridge is mounted on a 60 tape transport and automatically threaded and positioned for information transfer Each of six storage units has a capacity for storing a maximum of 80 cartridges and each cartridge handler serves two tape transports The instantaneous information-transfer rate of a tape transport is 1 128 000 8-bit characters per second This rate is based upon a packing density of 3 000 bits per inch on 1 75-inch-wide tape 22 tracks across 16 data bits 6 check-bits and tape motion at 2 35 inches per second The total system capacity that is the volume of data automatically available from storage maximum of 480 cartridges is approximately 44 billioni characters Figure 23 shows the TRACTOR automatic cartridge handling units in the left background and Figure 29 is a close-up view of one TRACTOR cartridge handling unit HARVEST OPERATIONAL SYSTEM HOPS When in May 1957 IBM submitted their HARVEST design proposal some of their personnel time was released from RANCHO the system design study contract However because both NSA and IBM recognized that an automatic programming system was a prerequisite for successful operation of the HARVEST equipment the RANCHO study was extended and provision made for part of IBM's effort to be devoted to laying the foundation for such a system Later in September 1958 a continuati9n o _j hifLE fJortwasprovidedfor under _a _new task FARM BOY IBM personnel at first proposed to create a system of programming for HARVEST that would be in effect a consolidation of work done in connection with the AEC STRETCH system with FORTRAN2 together with necessary additions to satisfy other needs Consideration of the effects of such an approach produced the decision that more emphasis was required on data-processing operations and especially on particularly analytic operations Meantime the need for definition of the kind of system required the classes of data to be dealt with and specialized operations to be performed led to establishment of the 1 Because of bad spots interrecord gaps and parity characters this approximation amounts on the average to 50 percent of the usable tape space 2 A widely used programming language created in 1956 by IBM 61 task CORN CRIB In July 1 59 a newly-established staff group began work on -a programming system for HARVEST The initial effort in this new group resulted in creation of specifications for an advanced language The newly specified language was called ALPHA Advanced Language for Programming HARVEST - In February 1960 in an attempt to reach a satisfactory basis for a division of effort NSA and IBM specialists met in conference The following decisions were reached 1 Because many ALPHA features were too sophisticated to be implemented in time for the delivery of HARVEST a minimum system should be created 2 The writing of the specifications for the various parts of the programming system should be apportioned as follows IBM Machine· Control Program MCP HARVEST Assembly Program HAP Executive or HARVEST Control Program HCP NSA Generalized File Operators GFO Job Request Language JRL In May 1960 NSA's part in this arrangement had been completed but because of personnel and training difficulties IBM was not ready with the HCP and MCP specifications The decision was made that NSA···should ··continue with the specifications of HCP leaving MCP and HAP with IBM By August 1960 about 35 IBM personnel were implementing ALPHA using as a basis the NSA-produced ALPHA Language Technical Report that was issued in June 1960 and the specifications for HAP were accepted In March 1961 a new task order CLAGHORN was signed that provided for the implementation of the HARVEST Assembly Program the Machine Control Program and the ALPHA Compiler also by IBM The total contract cost ' inclUding all previous FARM BOY costs was to be $2 lB9 000 and the work was to be completed by July 1963 CLAGHORN was later extended until 30 November 1963 The actual production of the programming system was a joint NSA-IBM effort with many of the decisions and conventions arrived at in conference and with subsequent production of different portions of the work assigned to one or the other of the two working geoups A HARVEST program is ordinarily written using one of four programming languages that occupy different hierarchical levels within HOPS To some extent statements from more than one language can be intermixed in a program These languages listed in order from the lowest close to machine language to the highest close to problem language are 62 HAP HARVEST Assembly Program Autocoder Macro-instructions ALPHA Advanced Language for Programming HARVEST JRL Job Request Language Because HAP really uses a system of mnemonics for the HARVEST instructions and merely converts these symbols into binary form on a one-for-one basis such programs are potentially the most efficient but they are also the most tedious to prepare In fact no practical programs are written using HAP alone Autocoder the next level of language above HAP uses a system of statements of common logical functions macro-statements A language-processing program converts these statements into symbolic form and then just as for HAP into binary machine language In practice the combination of HAP and Autocoder called HAP III is so commonly used that it is considered a single level of language In ALPHA the highest level for preparing a HARVEST program the programmer uses English words in referring to a sequence of logical operations upon data described as logical units of information ALPHA has been designed to create programs very rapidly compared with HAP and Autocoder and in addition can include statements in these languages The ALPHA compiler translates the programmer's statements into machine language going through a succession of passes Finally requests for jobs on HARVEST must use a pro 'cedure-oriented external command language -' 1Rt ' ----rn-fne--J6b Request Language the requester directs the entire HOPS complex including the HOPS Control Program HCP input and output functions data specification library of programs and individual functions in HAP or ALPHA The actions called for in JRL are carried out by a series of control programs that form the foundation for HOPS The philosophy of HOPS is based upon an operational cycle that processes work as much as possible in batches A HOPS cycle a collection of job requests varies according to the amount of wo k and volume of data involved The TRACTOR Tape System plays an integral role in HOPS operation All data files system control programs and problem programs and procedures are stored on TRACTOR tapes To facilitate problem programming for the job requester the data involved in each problem are organized into sets of logical records called logical files that can be referenced by name independently of detailed quantities and locations The manipulation of files including specification of individual subsets of logical files and maintenance of records and catalogs is done by a pa t of HOPS called TRACTOR File Maintenance Input to the whole system is performed by way of IBM Type 729 high-density tapes that contain blocked files prepared on a 63 -' 1401 computer in the HARVEST complex and are loaded through HARVEST onto TRACTOR tapes The output function is handled in similar fashion Output data is structured by the HARVEST program before recording on 729 tapes After removing such tapes printing of results if required is accomplished by using the 1401 with a line printer All problem programs available to the job requester are either generalized file operators GFO's or special-purpose programs GFO's perform most of the common data processing functions and each has available a number of variations in parameter specification As the name indicates special-purpose programs are written for operations too specialized for GFO's Additions to both classes of programs will continue to be made as the HOPS system grows In summation HARVEST as a hardware system is probably the most logically sophisticated computer in existence and even though based upon the 1956-57 state of the engineering art it is also one of the fastest computers in operation Because of the HARVEST system's logical complexity the HOPS effort itself a massive undertaking has imposed certain conventions to enable large numbers of programmers with a small amount of training to prepare jobs for HARVEST that can deliver results within a relatively short total elapsed time Of course sustaining programs that were written during the early HARVEST experience can be refined if the additional effort is justified Finally the whole operational programming s yatem __ is designed to be open-ended and even though the HARVEST complex has not yet been used to its fullest potential HOPS will grow and improve as the individual sponsors and programmers gain experience 64 CHAPTER 3 COMMERCIAL COMPUTERS As the commercial computer industry grew and requirements became better defined more computers became available to attack profitably specific types of Agency problems Commerc al computers grouped according' to the builder and ranging from the simplest desk-sized equipments to the most sophisticated machines are discussed in the following pages Several of these machines are better adapted for computation with a small quantity of input and output others are designed to handle large volumes of data using little computation and some are particularly well-suited for use in combination with other computers or with special-purpose attachments Those used as remote-operated systems are treated separately in Chapter 4 INTERNATIONAL BUSIN SS MACHINES CORPORATION IBM-701 When in 1952 IBM announced plans for building twenty Defense--Calcu-lators 'NSA--was-among the first to place an order This action was natural because of the Agency's long and successful use of IBM punched-card equipment The Defense Calculator renamed the EDPM Type 701 featured binary arithmetic and one-address instruction logic two instructions per word Each word of its high-speed 4 096-word electrostatic-tube-type memory was 36 bits long The 701 also included a slower drum memory of 16 384 words and card and magnetic-tape input-output devices In later 701 models magnetic core memories were substituted for the electrostatic tubes The 701 was delivered in April 1953 Many operational jobs being run on standard punched-card equipment or on ATLAS or ABNER were programmed for the 701 to provide flexibility in choice of available equipment Operating experience with the 701 revealed a number of weaknesses in the system as well as inadequate Agency preparation of programs and operating procedures First because no off-line card-to-tape preparation facility existed much valuable computer time was consumed during card input of data Operations involving large-scale use of the magnetic tape were plagued by failures because of faulty reading or recording on magnetic tape Humidity control in the machine area was critical but whether magnetic-tape errors were primarily due to this or to undependable read-write electronics would be difficult to determine However there were some instances of successful operations of large jobs some involving several hundred reels of magnetic tape The overall operations of 70l's internal computation on problems requiring limited data input or minimizing magnetic-tape operations was excellent The 701 was retired in December 1955 IBM-702 Soon after the 701 was operational IBM announced a new line of data-processing equipment The EDPM 702 used a binary-coded decimal rather than a pure binary number system in its arithmetic unit and operated upon variable-length fields rather than words It was planned with emphasis upon commercial and accounting operations involving large streams or lists of numerical or alphabetic material NSA's machine processing people recognized its applicability for its large-volume indexing and data-manipulation requirements and placed an order The 702 was delivered in the spring of 1955 The machine processing organization created a separate group to prepare for programming and operation of the 702 This group undertook the preparation of 702 programs to replace most of the standard-punched card-pro eduresfor largevolume data-handling jobs Preparation of indexes large-scale data conversion various types of speciali ed _dictionarypreparation are a few examples of large j obsfor which the 702 was considered suitable The 702 was retired in April 1956 IBM-705 Within a year after the installation of the 702 IBM announced a new machine the Type 705 as a replacement for the 702 The 705 was an improvement over the 702 because it had a large core memory a faster arithmetic unit and was generally better engineered In addition it had the ability to operate input-output equipment simultaneously with computation Requirements had increased to such an extent that the acquisition of five of these machines seemed feasible These were delivered one in May 1956 one in January 1957 two in September 1957 and one in February 1961 The first two were returned to IBM in November 1962 and April 1960 respectively Retirement of the remaining three 705's begins in March 1964 The 705's have made an impressive record of useful service in the many jobs included in the class of data processing D'ARTAGNAN generalized four-way sorting is particularly worth mentioning This program regarded as one of the most efficient sorting programs has been made available to other agencies Also the 705 installation made the first large-scale practical use of TRANSEMBLF R a procedure whereby large numbers of full-scale programs were assembled on magnetic tape for systematic processing Figure 31 is a general view of the 705 and Figure 32 shows the 705 core memory Because of the great number of IBM 727 magnetic-tape drives associated with a multi-70S installation a switching system was proposed that would make connection possible between any tape drives and any 705 tape control units Fewer tape drives would be needed with such a system thus possibly balancing the cost Because no such switching system was available commercially the RID group was requested to design and construct one SINBAD was completed in November 1957 It made possible the switching between a maximum of 15 Type 727 magnetic-tape units and the control units for the printers card readers card punches and tape drives associated with the several IBM 705 data-processing systems In Figure 31 the set of manual switches and controls for SINBAD can be seen just to the right of the console IBM-704 Coinciding with the introduction of the 70S IBM announced its new Type 704 as a replacement for the 701 Al _though_ the 704 had the sarne basic--logical organizat·iQn as the 701 its arithmetic unit and core storage were faster and its design provided additional features such as index registers and logical arithmetic The first 704 system was delivered to the Agency in January 1956 and was operational in one week The second 704 was delivered in February 1957 Both were equipped with 4 096-word core storage The third system arrived in September 1957 and contained' 32 768 W rds of core storage Subsequently the storage size of the first two systems was increased a so -- to 32 768 words The 704 systems proved to be much more reliable than the 701 as well as superior in capacity speed and logic Also operations involving much magnetic-tape manipulation were more practical because tapes were interchangeable among the 704's 70S's and off-line conversion equipments The first 704 was retired in August 1962 and the second and third in July 1963 Figure 33 is a general view-of the 704 and Figure 34 shows the 704 core memory IBM-650 ' The IBM-GSO a medium-speed computer with magnetic drum memory was delivered in December 1958 Its memory was sup67 'i' lur0 31 -- IHM 705 SIHJv'Jllij lN13f1U cit Ri Jht of 70S c JI ol t JlIL lO 1 i· JI'I l · P Figure 32-IBM-705 Core Memory 9f ' - -s ih ' - - 69 -- J -- --' Figure 3 2 -IBM-704 • • h f - - 70 Figure 34 -IBM-704 Core Memory plemented by a large random-access disk storage the RAMAC It used two-address instruction logic and a decimal arithmetic system each word contained 10 decimal digits plus sign The primary application planned for the IBM-6S0 was a specialized 'use of table look-up For this particular use it was well suited -- its disk storage had a capacity sufficient to store large tables Also the 650 was connected to as many as 10 consoles available to user groups to facilitate easy interrogation on a remote-operated basis In January 1962 the IBM-6S0 was retired Its replacement for this principal function the IBM-14l0 was installed in March 1962 Figures 3S 36 37 and 38 are views of the 650 IBM-1401 In December 1960 the first of a series of IBM-1401's was installed These machines were obtainGd as a generalpurpose answer to the perennial data-conversion and formatting requirements As mentioned in Chapter 1 no small part of any machine installation goes into the conversion function and many machines for this purpose have been built see alpo BOGART page 31 The 1401 Figure 39 is a transistorized machine similar to the IBM-70S in logic with somewhat greater flexibility and faster output yet it rents for less The second and third 1401' s were delivered in March and August of- 196 1 A total of 13 machines of this type have been installed to date The first two were returned to IBM in July and August 1962 in favor of other models with differing peripheral equipment configuration Two are assigned specifically for HARVEST peripheral support off-line magnetic-tape preparation and output printing and another is used as a satellite computer with the 7090 Others are used for paper-tape or card conversion and also exclusively as off-line printers of magnetic-tape recorded data One 1401 card input has been in use in connection with communications security since September 1961 in conjunction with a RAMAC disk memory IBM-1410 The 1410 is similar to the 1401 but includes a more flexible Move instruction and more index registers as well as faster internal memory It was installed in March 1962 primarily as a replacement for the IBM-6S0 It was originally equipped with four disk-memory modules with a total capacity of 80 million characters in addition to the 40 000character internal core memory The disk-memory modules were released in December 1963 and the corresponding work was taken over by magnetic tapes At the present time the 1410 71 0 _ It Lo'l 0 0 - - I I • 1 Figure 35-IBM-650 General View • f _ 0 ·'iiIJii - --- ntt 72 Figure 36 -IBM-650 Console _ - $5on $9 cmm mm nm 23E a 4 fr ma Figure 38-1131 650 Inquiry Station 9L 1 Mi Figure 39 IBM 1401 is being used primarily for data processing jobs formerly done on the 705 and punched-card equipment IBM-7090 IBM's successor to the 704 was the Type 709 Shortly after the 709 was announced the Agency placed a tentative order for one as a replacement for the 704 But because of budget limitations the order was cancelled Subsequently IBM decided to market the 7090 which is transistorized and embodies several other points superior to the 709 Orders were placed by this Agency for the 7090 and the first two were installed in July 1962 and January 1963 A third machine was delivered in November 1963 and a 1401 was installed to operate as a satellite computer with it IBM has developed and made available with the 7090 several programming packages including compiler report generator J input-output control and 704 simulation package I Whereas the 704 basic cycle-time was 12 microseconds the 7090 cycle-time is 2 18 microseconds The estimated speed superiority of the 7090 over the 704 is approximately in the same ratio Several other improvements facilitate operations requiring much magnetic-tape manipulation These include greater information density on tapes and several buffer channels for simultaneous reading writing and computing All three machines ar u ed 24_hours a day with most of the-use emphasizing mathematical and scientific jobs rather than dataprocessing The majority of the programs run'on the 7090's since their arrival were converted from those originally written for the 704 using the 704 simulation program Naturally these are less efficient than programs written expressly for t he 7090 Figure 40 is a view of the 7090 console GENERAL PRECISION J INC LGP-30 In the summer of 1956 a proposal was made to procure a computer for use primarily on mathematical problems The LGP-30 made by Librascope Incorporated was selected This computer is used mainly to test new ideas for analysis preliminary to requests for full-sized operations In addition it is used to perform complete jobs such as modest statistical tests suited to its capabilities For such applications much time is aved because the LGP-30 operating group is practically self-sufficient -- data is punched on tape a qualified programmer is available full time and the computer is 76 • - ' - - - -' - 'o ' ---- ' - t Figure 40 -- laM 7090 Console extremely reliable The LGP-30 is a desk-sized small-scale computer with slow-speed punch-tape and Flexowriter input-output and 4 096word drum memory of 7 5 milliseconds average access time Its instruction logic is one-address type one instruction per word Its internal -number system is binary and its wordsize is 30 bits plus sign In JUly 1958 the equipment was replaced by a new computer of the same type at no charge to the Agency because of major malfunctioning The replacement computer has been kept in an air-cooled room and has given extremely reliable service Figure 41 is a view of the LGP-30 CONTROL DATA CORPORATION CDC-1604 In certain circumstances the most effective use of a computer is made when it is reserved for full-time use on a specific problem This specialization is still more effective when a particular attachment is created that increases the efficiency of the computer for the problem NSA's first use of a general-purpose computer in this way began in September 1960 when a 'CDC-1604 was delivered with a special --attachment The complete assembly is caITedWELCHE-R _ As used in this particular problem application WELCHER is a more powerful tool than any available general-purpose computer yet the CDC-1604 retains its normal general-purpose versatility The combination is also undoubtedly more economical than a complete special-purpose machine Figure 42 is a view of WELCHER showing the digital display equipment DD-51 that operates directly in the system The CDC-1604 Figure 43 is a pow rful computer having a core memory of 32 768 48-bit words and an effective access time of 4 8 microseconds Its many flexible features include six index registers indirect addressing logical and masking operations program interrupt and an automatic storage searching capability as well as fixed and floating-point arithmetic Its input-output features include a paper-tape reader a punch and three input and three output buffer channels' thus permitting simultaneous inputting outputting and computing The CDC-l604 is also equipped with three IBM Type 727 magnetic-tape drives A high-speed transfer channel facilitates the operation of special-purpose attachments Two other CDC-1604 computers were delivered -- CDC-1604 2 in February 1961 and CDC-1604 3 in March 1962 These 78 L'w - -p vo-v 93 21 U 313% q uan a Er Mu Figure 4 18 Figure 43 machines have each been equipped with eight IBM Type 729-IV magnetic-tape drives and in addition the CDC-1604 2 has a special data display device the 00-51 Except for this change and their lack of special attachments they are identical to the 1604 of WELCHER PULLMAN installed in January 1963 includes a fourth CDC-1604 and a special-purpose attachment It is equipped with four IBM Type 729-rv magnetic-tape drives It is also equipped with the special data display device the 00-51 The two computers l604 2 and 1604 3 are further equipped with an additional powerful instruction the 1t77 Order By selecting suitable patterns of ones and zeros tremendous numbers of variations of transmissive and arithmetic operations can be called into play involving COC-1604 arithmetic registers and index registers This instruction increases their flexibility for analytic applications The increasing demand by research personnel for computer time and the difficulty of their competing with operations users has resulted in the establishment of a special research computing facility The CDC-1604A was selected and in July 1963 the Agency's fifth COC-1604 was installed It has a l OOO-line-per-minute printer a 'card reader and 10 magnetic-tape drives The research computing facility is operated as an open-shop operation as far as programitfirigis concerned- - arid an interesting arrangement whereby users can deliver programs by tube or by messenger has been evolved A supervisor program has been refined so that it not only compiles programs written either in FORTRAN or ALGOL but regulates true concurrent input-output tape-to-printer and internal computation Also a unique feature of this facility is the so-called bidder system for regulating priorities and maintaining records of machine time assignments According to this system individual users actually bid for m chine time using their own evaluation in relation to their overall monthly allocation The machine itself maintains the system records and issues reports to users and supervisors Overall system operation stresses quick turn-around Normally users can expect a maximum delay of 30 minutes from the submission of a program to delivery of the result including debugging jobs CDC-160A The CDC-l60A Figure 44 is a desk-sized general-purpose computer in a price range low enough to be practical for many data-conversion uses and yet sufficiently powerful and flexible for use in real-time and control applications It uses l2-bit words that are available in groups of 8 192 up to a max82 - In v bf Figure 44 Control Data Computer 83 imum of 32 768 word3 Typically the instruction word consists of a 6-bit function code and 6-bit execution address Provisions for specifying more than 64 individual addresses include the use of modifying registers that make available two additional modes of addressing -- indirect and relative Also available are several other rather specialized modes of addressing The storage cycle-time is 6 4 microseconds One hundred and thirty flexible instructions are availableithese make the CDC-160A very versatile for data-manipulation The machine is equipped with a high-speed paper-tape reader and punch a buffered input-output channel and an interrupt feature 84 CHAPTER 4 REMOTE-OPERATED COMPUTERS For years analysts have yearned for desk-top or deskside machines that could relieve some of the drudgery that always accompanies creative activity In 1942 the Navy had built a number of copies including several for the Army of the NCR Calculator a desk-top adding-machine-like device that facilitated tests requiring the use of non-carrying addition Other analyst-aids have been built for other purposes These together with desk calculators and various types of reference tables just about summarize the mechanical help that analysts had available for their personal use Attempts to build more elaborate equipment for use in analysts' areas were usually discouraged because of expense excessive noise or difficulty in maintenance The satisfaction of the analyst's need for personal machine support without many of the disadvantages usually associated with operating a computer installation finally took the form of a remote-operated computer one of the first in the country Three such projects are discussed A fourth the IBM-GSO was also used remotely but its use was limited to only one class of interrogations it has been discussed separately page 71 ROGUE ALWAC IIIE In February 1954 investigations were started by the research organization with the assistance of operational personnel to determine the feasibility of satisfying the need for more sophisticated machine assistance for analysts by a system of remote operation Under a research task AUTOMAT two approaches were studied The first was TOTALIZER a limited-purpose device designed to make comparisons counts and calculations of relevant stati'stics from streams of data on punched tape l The second ROGUE Remotely Operated GeneralUse quipment proposed to use a small-scale general-purpose 1 An engineering model of TOTALIZER was constructed but in October 1957 before its design was completed for outstation use the machine was placed in operation It was used profitably for some months but never as remote-operated equip mente digital computer to obtain much greater versatility The studies under the subtask ROGUE led to consideration of many commercial small-scale computers and a modified version of ABNER A report of these studies was prepared in August 1954 that recommended an experimental installation using the ALWAC IIIE with three remote stations The proposal was approved and in May 1955 a contract was signed with Logistics Research Incorporated Redondo Beach California providing for the delivery and installation of one ALWAC IIlE computer The contract was amended in December 19S5 to provide for a fourth station The machine was delivered in March 1956 and in April began regular operation r- The ALWAC IIIE computer Figure 45 used one-address logic had 8 192 32-bit words of drum memory only 128 words directly addressable and a system of relative addressing and block transfers Also in the basic system was a provision for cycling holding and indicator lights for accommodation of the several remote stations The speed of most operations was dependent upon placement on the drum and upon need for block transfers maximum access to words on the drum was 8 milliseconds 1 millisecond optimum and the average time for block transfer between main memory and working storage was 3 milliseconds per word ating Among outstanding features of the experience of oper been ALWA ha e 1 The limitation of individual station use normally to lS-minute runs thus keeping waiting time down 2 The total absence of administrative control or record keeping 3 The fact that a surprising variety of useful jobs could be preprogramrned by local analysts trained to program their own work so that useful results could be produced within the l5- inute limit In spite of certain limitations of ALWAC the experiment was an outstanding success in that analysts in operating areas programmed their own problems and utilized the system to advantage for small jobs Probably the most important effect of using ALWAC was to get the analysts into the act overcoming antipathy towards machines among many people with a resulting increase in rapport between analysts and professional machine people In September 1959 the use of ALWAC as a remote-operation system was discontinued to allow the factory to recondition the drum memory In February 1960 the computer was returned to operation without remote stations 86 LB Figure 45 ALWAC - ROB ROY BOGART In February 1957 another report was prepared by the research group that contained recommendations for a final ROGUE system Among the comments on the ALWAC operating experience the following disadvantages were cited 1 Slow speed 1 to 17 milliseconds per order 2 Limited addressable memory 128 words 3 F tremely slow input-output 10 char second However recognizing the overall success of the ALWAC experiment and the enthusiasm of the analytic groups the report recommended an improved system that included the following principal features 1 2 3 4 5 A BOGART computer A buffer system for magnetic tapes A buffer scan unit Auxiliary storage magnetic drums A high-speed paper-tape reader The report also recommended initiation of a development effort aimed at perfecting a practical high-speed output printer 200 characters per second In April 1 958 the research group was requested to proceed with the construction of a remote system substantially as described in the 1957 recommendations It was called ROB ROY The proposed system used BOGART Serial 5 as its central computer with a Ferranti high-speed paper-tape reader 200 frames per second at each of four outstations transmitting data to the central computer over telephone cables At the computer site the character rate is stepped up by means of a buffer system and written onto IBM Type 727 magnetic tape for computer input Output results follow a similar path with a 60-characters-per-second punch and an 0ff-line Flexowriter at each station to produce hard copies as desired The input and output core buffers and outstation consoles were constructed by Vitro Labs In January 1959 a fifth outstation was provided for and in December 1959 the ROB ROY system was installed and checked The installation at first had power line noise and long lead pick-up These difficulties were eventually eliminated by Agency engineers In February 1960 the ROB ROY System was placed in regular operation The total cost was $219 000 of which $152 500 covered the equipment furnished under the Vitro Labs contract The cost of the BOGART computer is not included in these figures The experience gained in operati q ROB ROY has been even more useful than that of ALWAC More sophisticated runs have been possible including fairly large analytic and dataprocessing jobs The speed of ROB ROY is greater than ALWAC's by a factor of several hundred The principal operating advantages over ALWAC are the following 1 Elimination of any elaborate sWitching l system for control of outstation operation The BOGART supervisor program calls for a program from the library tape assembles new programs if necessary and logs records of system operation 2 Actual effective overlap of input from outstations computer operation and output This is made possible by use of core buffers at input and output together with use of two magnetic tapes for temporary storage of data and results 3 Greater storage capacity and instruction flexibility of BOGART plus availability of three additional magnetic-tape drives for storage of intermediate results and for other purposes These add tremendously to the system's usefulness J The demand for ROB ROY has been much greater than its capacity to accommodate as a remote-operated system The resul t has been an arrangement -for many jobs -to----be- run nights and weekends In such cases there is an effort to batch jobs But the practical limitation in the number of outstations and the increasing demand for more and bigger runs have led to plans for a successor to ROB ROY Figure 46 is a view of a ROB ROY outstation and Figure 47 is a view of the ROB ROY control panel RYE UNIVAC 490 When the decision was made to 'plan a remote-operation system as a successor to ROB ROY bids from four possible contractors were solicited and their proposals were evaluated according to a set of stringent requirements The UNIVAC 490 Figure 48 was selected and orders for two systems were placed The first 490 was delivered in August 1963 the second is to come in January 1964 Both are planned to operate in a master-slave relationship and be fed by 30 remote stations 1 A scanner at input and a selector at output connect outstations 89 -- Figure 46 -ROB ROY Outstation -c _- - 4 ' 0 it' I '- • • '- · ' ' ' · 1 90 ' cP' 'ff lJ' ' -' t'igure 47-ROB ROY Control Panel 0- I6 Figure 48 490 RYE Squawk 1 uh eventually possibly 50 using two communication multiplexers and the lines of the existing secure telephone system The UNIVAC 490 system has the following features 1 Main memory of 32 000 30-bit words with 6-microsecond cycle time 2 Twelve Uniservo-IIIC tape drives IBM 729-compatible 3 One high-speed magnetic drum capacity 786 000 words average access time of 17 milliseconds 4 One Fastrand mass storage magnetic drum capacity 13 million words 5 One high-speed printer speed 600 lines minute 6 One modified Model 28 Teletype with c nsole keyboard and printer 7 One high-speed paper-tape reader and punch speeds 400 and 110 characters per second respectively 8 Various communication line terminals to accommodate various outstation requirements The philosophy of outstation operation presumes 24-hour 36S-day availability with all stations capable of inputting data and requests simultaneously and being serviced with a minimum of delay according to a priority system under automatic system control Three classes of outstations are planned differing primarily according to class of input output and display equipment A general station will consist of a Model 35 Teletype only a class II st atiori WIll-cilso have a BOSTIC or high-speed paper-tape reader and punch device class I stations will have Teletype and BOSTIC and in addition a UNIVAC 1004 line printer and other high-speed equipment as needed One station will ·be equipped with an x-y plotter connected to the 490 through a data-channel Programs will normally be limited to 30 minutes of running time An executive program REX regulates the operation of programs from the various stations RY the Agency's realtime program places programs in proper priority order taking into account actual assigned priority The system will keep operations records for all jobs permit interrupts for high priority jobs and supervise debugging aids to programmers Also a number of General utility Programs Guppies will be available to users of the RYE system The two 490 computers will be operable either together in master-slave relationship to provide extra capacity for extraordinary-capacity jobs or as separate computers In the latter situation the second 490 can be used for debugging or other operations while the first serves system needs If either 490 is out of operation the other can become the master and take over The RYE system is expected to become operational around 1 April 1964 92 APPENDIX r TABLE 1 CHRONOLOGICAL LISTING NSA GENERAL-PURPOSE ELECTRONIC DIGITAL COMPUTERS DECEMBER 1963 DATE RETIRED NOTES Dec 50 Nov 59 1 6 ABNER 1 Apr S2 Jan 58 1 14 RRU ATLAS 1 2 Mar 53 Nov 59 1 6 12 IBM 701 May 53 Dec 55 RRU ATLAS 11 1 Oct 53 Feb 60 1 13 RRU ATLAS 11 2 Dec 54 May 62 1 13 IBM 702 Apr 55 Apr 56 Techn ABNER 2 Jun 55 Jan 60 1 23 IBM 794 J _ Jan 56 Aug 62 7 67 3 8 85 BUILDER COMPUTER MODELl RRU ATLAS I 1 NSA DATE INSTALLED 2 PAGE 8 65 66 Log Res ALWAC IIIE Mar 56 IBM 705 1 May 56 Nov 62 66 IBM 705 2 Jan 57 Apr 60 66 Librasc LGP-30 Jan 57 IBM 704 2 Feb 57 RRU BOGART 1 Ju1 57 IBM 704 3 Sep 57 IBM 705 3 Sap 57 66 IBM 705 4 Sep 57 66 RRU BOGART 2 Nov 57 Ju1 63 9 76 7 67 1 31 Jul 63 67 1 33 1 Parenthes zed Arabic number after computer model designates 2 chronological order of delivery of a particular machine Please refer to NOTES foot of page 95 93 TABLE 1 Continued DATE INSTALLED DATE RETIRED NOTES PAGE Nov 57 1 33 BOGART 4 Jan 58 1 33 Phi1co SOLO Mar 58 1 2 29 IBM 650 Dec 58 RRU BOGART 5 ROB ROY Dec 59 1 3 33 88 CDC 1604 1 WELCHER Sep 60 4 78 IBM 1401 1 Dec 60 NSA CUB Jan 61 CDC 1601 Jan 61 82 CDC 1604 2 Feb 61 78 IBM 705 5 Feb 61 66 IBM ·1401 2 IBM 1401 3 Aug 61 71 IBM 1401 4 Aug 61 71 IBM 1401 5 Sep 61 5 71 IBM HARVEST Feb 62 1 39 IBM 1410 Mar 62 71 CDC 1604 3 Mar 62 78 IBM 1401 6 Jun 62 71 IBM 7090 1 Jul 62 76 IBM 1401 7 Ju1 62 71 IBM 1401 8 Aug 62 71 IBM 1401 9 Aug 62 71 IBM 1401 10 Oct 62 71 BUILDER COMPUTER MODEL RRU BOGART 3 RRU 94 - - _ Mar 61 Jan 62 67 Aug 62 71 1 36 71· - Ju1 62 TABLE 1 Continued DATE INSTALLED DATE RETIRED NOTES PAGE BUILDER COMPUTER MODEL IBM 1401 11 Nov 62 71 IBM 7090 2 Jan 63 76 CDC 1604 4 PULLMAN Jan 63 IBM 1401 12 Feb 63 RRU UNIVAC 1224A 1 CRISPI Jun 63 1 4 36 RRU UNIVAC 1224A 2 CRISPI Ju1 63 1 4 39 CDC 1604A 5 Ju1 63 RRU UNIVAC 490 1 RYE Aug 63 IBM 7090 3 Nov 63 76 IBM 1401 13 Nov 63 71 4 82 71 82 3 89 Builder Abbreviations RRU NSA IBM Techn Log Res Librasc Philco CDC Remington-Rand-Univac J Div of Sperry-Rand J Inc National Security Agency predecessor agencies International Business Machines Corporation Technitrol t Inc Logistics Research J Inc Librascope now Commercial Computer Div' J General Precision t Inc Phi1co Corporation subsidiary f Ford Motor Co Control Data Corporation NOTES 1 2 3 4 5 6 7 Agency-developed or sponsored computer design Operational use discontinued Aug 63 now used for training Agency-sponsored remote-operational use Used with Agency-sponsored special attachment Used for communications security purposes Additional memory 4096 words installed in 1956 IBM 704 1 and 704 2 memory capacity increased to 32 768 words in Jan 62 and March 62 respectively 8 Remote-station use discontinued Sep 59 9 Replacement new LGP-30 installed July 58 at no cost 95 r TABLE 2 ATLAS I INSTRUCTION CODE Clear Add Clear AR Hold Add Clear Add from Q Clear Subtract Hold Add from Q Hold Subtract Transmit A to Q Vector Add Q Jump Fill Q Jump Substitute Digits Sign-Conditional Jump Absolute Clear Add Zero-Conditional Jump Absolute Hold Add Clear Logical Multiply Absolute Clear Subtract Hold Logical Multiply Absolute Hold Subtract Print 0 n y __ Shift A Left Print and Punch Shift Q Left Intermediate Stop Substitute Execution Address Optional Stop Split Clear Add Final Stop Split Hold Add Cleal Multiply Split Clear Subtract Hold Multiply Split Hold Subtract Divide Store A Optional Jump Store Q Pass Clear Add Plus One Random Jump r- 96 TABLE 3 -- u s ELECTRO IC COMPUTER ACTIVITY IN 1947 EDVAC I I A S 2 UNIVAC RAYDAC Builder Moore School Univ Penna Institute for Advanced Study Eckert-Mauchly Computer Co Raytheon Electric Corp Ml'lll Ory Type Mercury Delay Electrostatic Tubes Mercury Delay Mercury Delay 25 130 400 Memory Access 200 microseconds Av Memory Capacity words 1 024 4 096 1 000 4 OBO WOrd Size 45 binary digits 40 binary digfts 12 decimal digits 45 binary digits Instruction Logic 4-address I-address I-address 4-address I EDVAC was the prototype for ABNER 2 The Institute for Advanced Study computer was built on an experimental non-commercial basis and was the prototype for ATLAS I as well as several other machines r TABLE 4 ABNER INSTRUCTION CODE STOP SWISH PROVISIONAL STOP SELECTIVE REPLACEMENT HALT PATTERN COUNT AUX ORDER SHIFT AND REPLACE TAPE BLOCK PENTABIT TRANSFER TAPE WORD EQUALITY PRINT UPPER EQUALITY COUNT PRINT LOWER LOW ORDER MULTIPLICATION COMPARISON MOD WITHIN WORDS COMPARISON COUNT SHORT MULTIPLICATION FULL MULTIPLICATION PENTABIT MOD EXPAND TRANSFER DIVISION RANDOM JUMP CYCLIC TRANSFER SKIP ADDITION FULL DIVISION DECABIT MOD SUBTRACTION 98 - REFERENCES Adams Associates Inc Computer Characteristics Quarterly Brummett Walter H Jr Atomic Energy Commission Report of Conferences in 17 Nov 5S Washington D C Regarding proposed Purchase of Computing Machine for LASL campaigne H H Report on Conference at 15 May 46 Navy Dept Survey of largescale automatic computing machines by J vonNeumann Control Data Corporation IDA CRD 77 Order for 1604 Preliminary Engineering Draft undated Control Data Corporation CDC l60A Computer Jul 63 Engineering Research Assoc Proposal for a Magnetic Core Memory for ATLAS I 19 Nov 54 1 Nov 63 2 16 Oct 63 House of Representatives Committee on Post Office and Civil Service Use of Electronic Data Processing Equipment in the Federal Government House of Representatives Committee on Post Office and Civil Service Inventory of Automatic Data 25 Oct 63 Processing Equipment in the Federal Government IBM Data Systems Division Final Report IBM 7950 Data 1 sept 62 IBM Data Systems Division Final Report Development 30 Nov 63 of an Automatic Programming System Processinq System 99