Software Development at the Eckert-Mauchly Computer Company Between 1947 and 1955

Iterations – Norberg – Software Development at EMCC 

 5 

keeping contact with potential customers, and overseeing the 

programming group, worked with several people on such problems as the 

generation of random numbers, bi-harmonic problems, and programs for 

BINAC. Jennings spent most of her energies on test routines for BINAC. 

She even used some time programming chess and gin rummy games.

9

  

 

Rounding out the early programming group, Grace Murray Hopper, who 

joined EMCC in 1949, possessed substantial training in mathematics and 

experience in computer development and coding. She studied mathematics 

at Vassar College in the 1920s, and in 1934 received a Ph.D. in 

mathematics from Yale. Hopper was elected to Phi Beta Kappa and Sigma 

Xi for her accomplishments. For a spell in the 1930s, she taught 

mathematics at Vassar and held a postdoctoral fellowship award from 

Vassar, which she used at New York University in 1941. She entered the 

US Navy in 1943, and, after training, was posted to the Bureau of 

Ordnance Computation Project at Harvard University. Working with 

Howard Aiken on the Mark I and Mark II, she developed several 

programs, and served as co-author of the Mark I and Mark II computer 

manuals.

10

 Hopper joined the Harvard staff as a Research Fellow in 1946, 

continuing to work on Mark II and Mark III for the navy. From this 

position, she moved to EMCC. Rounding out the early group was Herbert 

F. Mitchell, who joined EMCC in 1949 and became head of the new 

Laboratory for Computational Analysis. Mitchell received a Ph.D. from 

Harvard in 1948, where his dissertation advisor was Howard Aiken and he 

worked on developments in numerical analysis. Hopper and Mitchell had 

worked together on various Mark designs at Harvard after the war.  

 

The EMCC programming activities fell into three categories. First, there 

were the requests of the customers for programs to accomplish their tasks 

of payroll, procedural flows, accounting, and purchasing. Second, there 

was commercial research undertaken by EMCC to enhance sales of 

computer systems: collation, sorting, merging, editing, tabulated 

examples, and integrated systems for payroll. And third, EMCC engaged 

in engineering studies in collaboration with the engineering design group. 

Besides the logical design of BINAC and UNIVAC, they investigated the 

logical design of the UNITYPER, card-to-tape converter, supervisory 

control, speed comparisons of codes, error detection, and reliability.

11

  

 

To understand this group’s activity better, it is useful to return to March 

1947, an important month in EMCC, as the engineering and programming 

groups made many decisions about basic design, which no doubt is the 

reason Snyder served as a consultant during this time and the applications 

group began to grow in the following summer. For example, March 11 and 

12 were devoted to a design conference on the EDVAC II. During this 

conference the staff discussed the basic activities of their design—sorting 

speed, conversion of data from tape to memory and back, instructions for 

---

Iterations – Norberg – Software Development at EMCC 

 6 

operating the tape system, displaying the memory, instructions for moving 

from one point in a program to another, and starting and stopping. The 

give and take during the meeting, as reflected in summary minutes, 

resulted in some major decisions of how the storage system would be 

designed and operated.

12

 Census problems would require significant 

sorting, and if the EDVAC II used binary sorting instead of decimal, then 

gains over punch card systems would be offset making the EDVAC II and 

the punch card systems near equal for sorting. Moreover, in systems with a 

long latency period, sorting would be very time consuming. Times of 

recovery of data with tapes were estimated to be about ½ millisecond. “If 

mercury tanks holding 20 words are used, the average time lost in 

obtaining a word from such a memory is ½ millisec.” This meant that the 

internal speed of discriminating was comparable with the speed of the 

input and output.

13

 Calculation revealed that the gain in sorting using 

EDVAC II would be in the neighborhood of 2.5, a bad comparison when 

other tasks on the system showed gains of 100 to 1000. Some adjustments 

would have to be made. They discussed decreasing the latency time by 

using 10 word tanks instead of 20 word tanks. They believed they could 

insert some parallelism by providing for simultaneous use of internal 

operations and tape reading and writing. Data could then be transferred in 

blocks rather than individually.  The discussion at the meeting was done 

under an assumption of a pulse rate of 1 megacycle. In further 

consideration of the input/output issues, the group evaluated doubling the 

pulse rate and returning to the 20 word tanks, but this raised reliability 

questions. Further study seemed needed. Virtually all of these ideas were 

tried over the next year.  

 

Some of the instructions to accomplish these operations were obvious to 

the group, such as the arithmetic functions. Therefore, they only spent 

time considering additional orders to transfer data in the most efficacious 

manner. “It is desirable to make the orders used for reading and writing on 

tapes as simple as possible for the operator to use, and as simple as 

possible to ‘mechanize’ in control equipment.”

14

 Of four tapes, the system 

could run two at a time, with provision for running forward and reverse 

and reading and writing on different tapes simultaneously. Several 

instructions were to be designed for this purpose.

15

 Over the next six 

weeks, Mauchly and Snyder designed the first instruction set of 26 

instructions, five of which would later be dropped.

16

 

 

Between March 1947 and May 1949, the applications group developed 

and analyzed ten variations of the code, or instruction set, for EMCC 

designs and design changes. Many similarities exist across these codes, 

although there are some important differences. C-1 through C-4 were 

based on a 2-megacycle pulse repetition rate; the next four schemes 

involved a 4-megacycle rate. C-9 dropped back to 2 megacycles, and C-

10, the final instruction set for UNIVAC I, was to operate on 2.25 

---