Monday, October 01, 2012

Fact checking George Dyson (where he taps me on the shoulder)

It's no secret that I wasn't impressed by George Dyson's book "Turing's Cathedral" because it skewed history in a particular way a bit too much, and I felt that the title exploited the Turing anniversary. But I was struck by something he said in a StrataConf EU keynote.

He said that in 1953 there were only 53 kilobytes of random-access memory in computers in use and showed a picture of a February 1953 report entitled "A Survey of Automatic Digital Computers" published by the US Office of Naval Research. I thought that sounded odd, so I tracked down a copy of the report.

In fact, he makes the same claim in the book, but I'd overlooked it:

So, I started going through the report looking at machines that were operational in March 1953 according to the report. Just concentrating on binary machines I quickly found that random-access memory was well past 53KB.  By the time you reach E (the machines are in alphabetical order) there were 85,856 bytes of memory (in the ACE Pilot, APE(R)C, AVIDAC, BARK, EBIAC, CADAC, CSIRO Mark 1, EDSAC 1,  EDVAC, ELECOM 100 and ERA 1101).

So, then I wondered if, given that Dyson is oddly obsessed with the Williams Tube memory storage (making it the basis of his really odd "Turing Machines are one-dimensional, von Neumann machines two dimensional metaphor) if he was only counting machines that used fast electrostatic memory (he does, after all, say 'high-speed' in the book).

The machines that were operational in March 1953 with that type of memory are (according to the report): AVIDAC (5,120 bytes), ILLIAC (5,120 bytes), IAS Machine (5,120 bytes), Manchester Machine (645 bytes), MANIAC (5,120 bytes), ORDVAC (5,120 bytes), SEAC (2,816 bytes), SWAC (1,152 bytes), UTEC (768 bytes), Whirlwind (4,096 bytes).  That's a total of 35,077 bytes.

Reading the report it's clear that there was more than 53KB of memory in March 1953 and that it was spread across a variety of different memory types (magnetic drums, acoustic/mercury delay lines and electrostatic techniques).

It's a nice soundbite that in "1953 there were 53KB of memory", but like everything else in Dyson's book it's important to read it in the light of his fundamental idea that von Neumann's IAS machine is the Ur Machine.

PS After my talk at StrataConf I attended the FOO Party where all the speakers and others get together at the end of the conference. While chatting with someone, I felt a tap on my shoulder and turned round to hear "I really enjoyed your talk today. Marvellous to hear about LEO". The finger belonged to George Dyson.

So, of course, I had to own my opinions and ask him straight to his face about things I've written. I still don't understand his explanation of the one dimensional/two dimensional thing, but he convinced me that he's right about the 53KB in '53 thing. Apparently, there are two other important machines not mentioned in the US Navy report: one at IBM and another at RAND. He was just counting high-speed electrostatic memory as I'd guessed. And given that he spent 6 years researching the book he's probably got that part right!

No comments: