National Bureau of Standards, U.S.A.
A survey of automatic high speed digital computing in the U.S.A. with emphasis on the program of the National Bureau of Standards.
The first essentially electronic digital computer was the ENIAC constructed at the Moore School of Electrical Engineering of the University of Pennsylvania. This machine has been in operation (except for a short period while being moved to Aberdeen Proving Ground) since early in 1946. Although designed primarily for integrating systems of total differential equations (such as the exterior ballistic equations) the calculator has been used for general purpose computing. Types of problems that have been done on it include data reduction of trajectory information, solution of simultaneous linear equations, and the integration of partial differential equations.
When used as originally designed the procedure of changing from one problem to another involved the changing of several hundred pluggable connections and resetting an equivalent number of switches. During 1948 certain improvements on the ENIAC were made. One of the major improvements was a new system of coding. With this new method of coding, new problems can be put on the calculator without manual changing of connections or switches. This new system reduces the operating speed of the ENIAC, but increases its utility, due to the ease of changing from one problem to the next. If speed is at a premium and the problem is of sufficient size to merit it, the old system can still be used and the maximum speed of operation attained. With the development of this new coding system two new units were added to the ENIAC. One was a unit to facilitate the use of the new coding system. The other was a mercury delay line memory unit which would substantially increase the memory above that of the original memory of twenty ten-decimal digit numbers. This delay line unit was just being put into operation at the time of the writing of this article.
With the completion of the ENIAC the group at the Moore School started design work on a mercury delay line computer known as the EDVAC (Electronic Discrete Variable Computer). In the spring of 1946 Mr. Eckert and Dr. Mauchly (the persons primarily responsible for the design and construction of the ENIAC) left the Moore School and started their own company. The work on the EDVAC at the Moore School was carried on under the leadership of Mr. T.K. Sharpless. After a while Mr. Sharpless resigned from the Moore School and went into business for himself. Mr. Eckert and Dr. Mauchly planned to design and build large scale high speed digital computers, whereas Mr. Sharpless was more interested in developing and marketing computer components. Work on the EDVAC at the Moore School was carried on under the direction of Mr. Snyder.
As of the present writing the EDVAC is completely fabricated but has not been in operation as a whole. Most of the units have been checked and operated in a satisfactory manner. The last units now being worked on are the magnetic wire input-output units. Soon (before final check-out) the EDVAC is to be moved to Aberdeen Proving Ground. There it will be put into operation.
In 1946 a high speed calculator project was started at the Institute for Advanced Study under the direction of Dr. J. von Neumann. This machine has come to be unofficially known as the MANIAC. Originally it was to use the Selectron tube, under development at Radio Corporation of America (RCA), to provide a quick access electrostatic memory. Delays in the development of the Selectron tube led to magnetic drum investigations, and finally, to investigation of the use of standard cathode ray tubes in the manner developed by Professor F.C. Williams of Manchester University, England. Since then R.C.A. has developed a Selectron tube capable of storing 256 points Which seems to work satisfactorily. At the time of the writing of this paper it is still uncertain which of the three methods of storage will be used in the memory of the Institute of Advanced Study Electronic Computer. The group at the Institute of Advanced Study has constructed an arithmetic unit, and are now testing it. A magnetic wire handling device has also been developed by them.
In 1946 the Census Bureau became interested in the use of high speed machines for the tabulation of Census data. They transferred, funds to the Applied Mathematics Laboratories of the National Bureau of Standards for the purpose of obtaining a high speed calculator. At approximately the same time, certain other funds were transferred to the Bureau for the purpose of developing electronic calculating machines for other government agencies. It was decided to give development contracts to two companies. The purposes of these contracts were to obtain design plans and to pay for the construction of certain components to show the feasibility of building mercury delay line calculators. These contracts were given to the Eckert Mauchly Computer Corporation (then called the Electronic Control Co.) and to the Raytheon Manufacturing Corporation. As a result of these contracts design plans were produced, typical memory units were designed and operated in a satisfactory manner, and certain magnetic tape handling equipment was built. (It is of interest to note that the memory unit of the Eckert Mauchly Corporation operated at a pulse repetition frequency of five megacycles.)
These contracts were terminated early in 1948 and after evaluating the proposed designs the Bureau contracted for three calculators from the Eckert Mauchly Computer Corporation and one calculator from the Raytheon Corporation. These machines are expected to be delivered late in 1950 and early in 1951.
Meanwhile the Eckert Mauchly Computer Corporation received a contract from Northrop Aircraft Corporation to construct a 500-word memory binary calculator called the BINAC. This machine is going through final check-out prior to delivery. It has operated without failure for as much as forty hours at a time.
Due to the late delivery dates the National Bureau of Standards decided in June, 1948 to build an interim computer in the Electronics Division in Washington, D, C. To facilitate construction, this machine was copied after the EDVAC but on a simplified scale. Fabrication of the final chassis for this machine is now in progress.
In 1948 the National Bureau of Standards gave a design plan contract to Engineering Research Associates of St. Paul, Minnesota, and Washington, D.C. The purpose was to design a magnetic drum computer and to indicate techniques or circuitry by use of which the machine could be constructed. The results of this contract were available early in 1949. A magnetic drum computer with a memory of 16,000 words was proposed by Engineering Research Associates (ERA).
In January, 1948, the International Business Machines Corporation unveiled an electronic calculator in New York City which is comparable to the ENIAC. This machine has an exceedingly slow speed memory in the form of punched paper tape (punched card stock) which passes various reading stations. Its high speed electronic memory holds eight words; it has an intermediate speed memory constructed of relays which holds 150 words. Basically, the machine is tape sequenced (reads its instructions from paper tape). However, the machine can obey instructions stored in its memory.
At Massachusetts Institute of Technology, Cambridge, Massachusetts, a project has been investigating computing machine components. Making use of the results of this investigation, they are now constructing an exceptionally high speed digital computer called Whirlwind I. The memory is to consist of storage tubes to give quick access, and the machine is to operate in parallel to give high speed. Whirlwind I, with the exception of the memory unit, is nearly complete. It is hoped that it will be completed in one year.
Harvard University, under the direction of Dr. Howard Aiken, has constructed two calculators and is now working on the third. Mark I and Mark II, as the first two machines are called, have IBM type mechanical counters and relays for memory, and are comparatively slow speed. Mark III will be unveiled in September. It is a magnetic drum calculator, and is tape sequenced.
At the end of 1948 the Bureau decided to build an electrostatic calculator at the Institute of Numerical Analysis in Los Angeles, California. A laboratory was set up and personnel were recruited. At the present time a group of twelve people are working on this project. A parallel arithmetic unit has been designed and a three digit prototype of the arithmetic unit has been constructed. The time for carry propagation in these three units has turned out to be 0.07 microseconds. This indicates that a 36 digit (binary) accumulator can be made which will have a net addition time of less than 3 microseconds. Therefore, it can have a net multiplication time of less than 150 microseconds. (Net times do not allow for access times to the memory for operands, results, or commands). A complete deflecting system for use with the cathode ray tube memory has been built, and is in the process of being tested. The logical design of the control unit and the input-output units is complete. The initial input-output units will be standard teletype equipment. This equipment is already on hand. The logical plan calls for incorporating a magnetic drum as one of the input-output units. This will actually serve as an auxiliary (and to some extent permanent) memory, partly to remember subroutines, and partly to store intermediate results in the computation processes to be done on the machine.
From the performance point of view of the automatic digital computing machines now under construction in the United States, the following facts may be mentioned. The fundamental pulse repetition frequencies range from about 125 kilocycles in the cathode ray tube machines and the magnetic drum machines to about four megacycles in the UNIVACS designed by Eckert and Mauchly.
The mercury delay line machines are all serial, that is, words are represented by a time sequence of pulses. The cathode ray tube machines and the Engineering Research Associates magnetic drum machine are parallel. Mark III is a serial four channel (serial in terms of decimal digits) calculator.
The number of digits per word ranges from 16 binary digits in Whirlwind I to 44 in the EDVAC and the NBS Interim, with 16 decimal digits for Mark III. Maximum access time from the memory for a particular word ranges from 16 microseconds or less for the storage tube machines, through 350 to 384 microseconds for the delay line machines to 8,000 microseconds for one of the magnetic drum machines.
The average addition time (total including access times for operands, result, and command) ranges from possibly as little as 20 microseconds for Whirlwind I, to between 700 and 1000 microseconds for the delay line machines. The time for performing a floating addition (handling two numbers in the form a × 2b or a × 10b ) ranges from less than 3 milliseconds for the computer being built at the Institute for Numerical Analysis, through 30 for the UNIVAC and 60 for the EDVAC, to 175 milliseconds on the NBS Interim.
The amount of built-in automatic checking ranges from none in several machines, to very elaborate and thorough checking in the Raytheon calculator.
The command systems used range from one-address types used in the MANIAC and UNIVAC to four-address systems (addresses of two operands, a result, and address of the next command) used in the EDVAC and Raytheon computers. The Institute for Numerical Analysis calculator uses a modified four-address system in which the address of the operands and the result are specified, but the address of the next command is determined by a counter in the same way as in the one-address system. The fourth address specifies the next command in exceptional cases such as overflow (when capacity of the accumulator is exceeded).
In summary, the ENIAC and IBM electronic calculator are both in operation at the present time. The BINAC (500-word memory) is going through final check-out at the Eckert Mauchly Computer Corporation, while the Mark III is scheduled for unveiling at Harvard University in September. As far as the large scale electronic automatic sequenced digital computers are concerned, it seems as if the EDVAC will be the first of these to be put into operation.
