7

On the Funny Things Seen In Source Code And Documentation page at the WikiWikiWeb I ran into the following:

Legend has it that the following is found in an ancient IBM manual:

  • Single bit errors are detected and corrected
  • Double bit errors are detected
  • Undetected errors are ignored

I doubt that something like this would make it into an official manual, and íf it's true, I'd love to see the context.

chicks
  • 219
  • 1
  • 2
  • 10
Daniël van den Berg
  • 179
  • 2
  • 10
    uhm, how can you react to an error you don't know about? Undetected things are always ignored, this is how our reality works. The manual is merely restating the obvious to exhaust all possibilities. – Agent_L Jun 30 '16 at 12:16
  • 11
    Bit more context, for the people who aren't au fait with computer network theory: It sounds like the manual was talking about a [SECDED Hamming Code](https://en.wikipedia.org/wiki/Hamming_code) which is a special number transmitted with data when computers communicate over a noisy channel. If there is corruption of a single bit of data, the receiver can work out which bit is wrong and correct it. If there is corruption of two bits of data, the receiver can detect it and request a retransmission. But if there is corruption of >=3 bits, it might be undetected, or it might be wrongly corrected. – Oddthinking Jun 30 '16 at 12:51
  • 3
    @Oddthinking Not that it matters for the question, but the ability to correct N errors and detect M errors, where M > N, is a generic feature of most block codes and not a specific feature for hamming codes. – Tor-Einar Jarnbjo Jun 30 '16 at 13:31
  • The context could be error-correcting memory, error-correcting storage, error-correcting communications, or even something else. The ability to correct an error in a single bit in a word or a block and detect two errors has been standard stuff in the computer industry for a long time. You have no protection if you have a typical desktop, laptop, or portable device. One single event upset and blam. The probability of this happening during the life of your machine is small (assuming you live within a few thousand meters of sea level). The probability is very significant for big data centers. – David Hammen Jun 30 '16 at 15:44
  • 2
    It sounds like a joke to me, but odd thinking's explanation makes it not so funny. –  Jul 02 '16 at 15:58
  • This is from the same people that amusingly made [mouse balls](http://www.neystadt.org/john/humor/IBM-Mouse-Balls.htm) famous so I wouldn't be surprised. – chicks Jul 06 '16 at 17:57
  • @DavidHammen, the odds of getting an error depend on the computer, the memory, the operating temperature, and many other things. In college, I ran a computer lab with parity-based memory in the workstations (capable of detecting but not correcting errors), and I'd see errors at a rate of about one per gigabyte per month. On the other hand, my current desktop with 64GB of RAM went 24 hours of Memtest without an error, so the error rate is probably well below that. – Mark Jul 11 '16 at 02:53
  • 1
    About 30 years ago I worked with IBM System/32 and System/34 computers. One of the manuals that IBM supplied contained a matrix showing how to relate employee performance to skill. It was a complete joke, with things like (going from best performance to worst), "talks to God", "talks to himself," "argues with himself" and finally "loses these arguments." This was all in a very dull technical manual, so yes, it's entirely possible that something like you mention could make it into an IBM manual! – Avrohom Yisroel Aug 08 '16 at 22:59

0 Answers0