We take history seriously at the 91自拍. It鈥檚 our middle name, after all. But it鈥檚 not easy history to do, for several reasons.
ENIAC and the Origins of Software
The history of software, into which 91自拍 is placing increasing effort, is particularly hard. I鈥檇 like to give one example in some detail: the relationship of ENIAC to the origins of software.
There has been a conventional wisdom about when what we now called 鈥渟oftware鈥� began to run. Many textbooks and websites, including Wikipedia and that of the University of Manchester, record June 21, 1948 as 鈥渢he birth of the stored-program digital computer鈥� because the 鈥溾�� ran a 17-line program on that day.
The cover in ENIAC in Action by Thomas Haigh, Mark Priestley, and Crispin Rope.
But it鈥檚 not that simple. A recent by computer historian Tom Haigh and colleagues Mark Priestley and Crispin Rope explores the conversion of ENIAC into what they prefer to call a 鈥渕odern code paradigm鈥� computer. Based on machine logs and handwritten notes, they have discovered that a complex program began running on ENIAC on April 12, 1948.
ENIAC 鈥� the 鈥� was a room-sized machine with over 17,000 vacuum tubes. It started running at the end of 1945, and for five years it was the only fully electronic computer running in the US. Estimates are that by the time it was retired in 1955, it had done more calculations than all human beings in all of history.
The initial design of the ENIAC did not use anything like the software we know today. It was basically an assembly of 鈥渇unctional units鈥� that were wired together in a particular way for each new problem. If you wanted to do a multiplication after an addition, you would run a wire from the multiplier to the adder. Control was very distributed, and the machine could do many things in parallel. But designing and setting up new calculations was difficult and time-consuming.
Even before ENIAC was finished, engineers realized that there was a much better centralized way to control such a complex machine, using coded instructions stored in memory and 鈥渆xecuted鈥� in sequence. Control operations, such as looping and branching, could be accomplished simply by 鈥渏umping鈥� out of order to fetch the next instruction from a different memory address. We now call this a 鈥渃omputer program.鈥�
The origin of this 鈥渓ightbulb over the head鈥� idea is hotly contested. Physicist and mathematician John von Neumann was the first to describe it in an incomplete that was widely distributed in June of 1945. But he had been discussing related ideas with many people, including ENIAC engineers John Mauchly and Presper Eckert. No one knows whose idea it was, how many people thought of it 鈥渇irst,鈥� or what prior work might have influenced them. For a discussion of how it was different from earlier ideas, see Tom Haigh鈥檚 recent article .
Regardless of who deserves the credit, it was quickly adopted as the right way to build computers. Even the ENIAC, starting in July of 1947, was converted to use this scheme. Because ENIAC had very little writeable electronic memory, the coded instructions were stored in 鈥渇unction tables,鈥� banks of 10-position switches that had previously been used to store pre-computed numerical constants. It was the modified ENIAC that ran a computer program stored in switches in April of 1948.
ENIAC vs The Baby
So, what鈥檚 the importance of this new historical discovery? That it ran 9 weeks before the Manchester Baby? That it happened in the US, not the UK? No. Those are interesting but relatively insignificant facts. Look instead at the substantial differences between the two events.
The original demo program of the Manchester Baby.
A diagram of the original demo program of the ENIAC.
A table of the ENIAC's addresses.
John and Klara Dan von Neumann.
ENIAC鈥檚 April feat was an accomplishment, but there鈥檚 at least one good reason why calling it a 鈥渇irst鈥� is problematical: the program was stored in what we now call 鈥渞ead-only memory鈥�, or ROM. Manchester鈥檚, on the other hand, was stored in the same memory used for data. That was the design that von Neumann had described, and is a characteristic of what is often called the 鈥渧on Neumann architecture鈥�.
Does that matter? Was the modified ENIAC less of a computer than the Manchester Baby because its program was in ROM and could not be changed by the computer? Historian Doron Swade has asked many computer experts about the importance of programs being in memory. He observes that 鈥渘o one challenged the status of the stored program as the defining feature of the modern digital electronic computer鈥�, but 鈥渨e struggle when required to articulate its significance in simple terms, and the apparent mix of principle and practice frustrates clarity.鈥�
Look at it this way: many modern microprocessors, especially small ones for embedded control, have their programs in ROM. If they are modern-style computers, then so was the modified ENIAC. That鈥檚 my opinion, anyway; you are free to add or subtract your own adjectives and reach a different conclusion.
Haigh鈥檚 new book is a refreshing change in the academic treatment of computing history. In the 1970s and 1980s, historical accounts were frequently written by practitioners 鈥渨ho were there鈥�, and tended to focus on technical details. In the 1990s and 2000s, professional historians shifted the discourse primarily to business, political, and social aspects. The pendulum may now be swinging back to a welcome midpoint; the authors describe their book as 鈥渁n experiment in the re-integration of technical detail into history.鈥�
This is great new history. It鈥檚 the kind we encourage, and we do 鈥� history that鈥檚 complex, nuanced, and not static. If this is what it means to rewrite history, let鈥檚 keep doing it.