As a result the census office ordered 56 machines at a cost of $56,000 per annum. Hollerith had to service the machines and pay a penalty of $10 a day if a machine didn't work. You can recognise in these arrangements the start of the standard practices of the mainframe industry that would follow!
An early Hollerith tabulator
The 1890 census was completed in just one year - which has to be compared to the eight years needed for the 1880 census was reckoned that Hollerith's machines saved $5 million or more. The census was the first ever large scale data processing operation. There is no doubt that the census organisers took advantage of the new equipment to ask more complex questions - so the savings could have been even greater! Perhaps this was also the first example of automatic data processing simply extending the job to fill the time available! It also started the reputation that data processing has for being boring. The workers employed to use the tabulating machines were so bored that they used to suck out the mercury with an eye dropper to gain a few hours of fun watching the machine being repaired! It was also the first large scale employment of women as office workers.
The Tabulating Machine Company
After the success of the census Hollerith machines were adopted by business in general. First US railroad companies adopted it - but only after a great deal of effort on Hollerith's part. He set up a business called The Tabulating Machine Company - although he didn't get round to setting it up properly for some time.
First he had to change the card system to accommodate the railroad data. This time he chose a card that was the same size as the current dollar bill - a size that remained in use thereafter. He also introduced a system of fields - sets of columns that recorded a single item of data. The cards were printed with rulings and now could be read by people as well as machines.
The beginnings of a full data processing system was forming and it was common to refer to it as the Hollerith System and the punch card was the Hollerith Card. Over the years the system developed in sophistication. The electromechanical machines to handle cards became faster and more capable. Soon after the basic counters and sorters came machines that could add up the contents of fields and even multiply values. A punch card data processing centre was the forerunner of the mainframe installation. The cards were a basic form of memory, the card machines formed the elements of a processor.
The only missing component needed to call the system a computer was a program that controlled the entire machine. Instead operators had to decide what had to be done with the stacks of cards and manually move them from machine to machine. If you wanted to, you could argue that the people running the machines were the program running the first automatic data processing machine. So only fifty years after Babbage had given up on his difference and analytical engines, technology had improved to the point where automatic punch card data processing was possible, useful and profitable.
Having succeeded with his invention, Hollerith resisted new ideas for the operation of his machines with dire consequences. Around 1905, the U.S. Census Bureau gave him an ultimatum: improve the machines and cut the rentals. Hollerith refused and the Census Bureau carried out its threat of building the enhanced machines which were then operated by former employees.
Hollerith sued the government for infringing his patents and wrote irate letters to newspapers and to the President. He finally lost the legal argument in 1912.
By that time Hollerith had sold the Tabulating Machine Company and it merged with a group of other companies to become Computer-Tabulating-Recording Company. In 1914 CTR hired a salesman, Thomas J Watson as general manager. Not only did Watson set about to motivate the salesforce, he also called for an improved machine that would print results automatically. Hollerith refused to have anything to do with the resulting machine and although he stayed with the company until 1921, he avoided Watson as much as possible.
In 1924 the company was renamed International Business Machines (IBM) and so opened another chapter in the history of computing.
Hollerith died in 1929 but his punch card system was in use well into the 1970s and later. Today the only time anyone remembers his name is when a programmer meets an oddly named text format in FORTRAN - the Hollerith field.
Donald Knuth has been described as the Euclid of computer science. The first draft of his epic "The Art of Computer Programming" was completed as a 12-chapter manuscript in 1965. Fifty years later TAO [ ... ]