hdrw@ibmpcug.co.uk (Howard Winter) (04/12/91)
Memory doesn't always fail 'hard' by any means, and the power-on test only detects hard failures. Back in the mid 70's I had a Z80 based machine which I played with, and it developed a single-bit fault that took between 5 and 15 seconds to happen - you could set a value in a mmemory location and then run a display loop which showed the value. Suddenly, with no provocation the value of a single bit would change. Set it back and 'verify' it - OK. A few seconds later it changed back again. I isolated the chip, and changed it - all was well after that. It seemed a bit of a shame to throw away a whole chip for a single-bit fault, but then most of the Titanic didn't leak... This system didn't have parity checking, so if I hadn't known about the fault, the effect could have been an error occurring in data (and being written back to disk) would have gradually corrupted a database, and in program code could have caused almost any unwanted side effect - just look at the difference changing a bit in a machine-code instruction makes... I thI think memory parity checking is *vital* for any serious equipment - certainly for commercial machines, and in my case, for my home machine. I am a little surprised that with 32-bit machines that simple 1-per-byte parity is used instead of more sophisticated correction, but there you are. I don't know if 4-per-32 can do any correction - perhaps someone else out there can say ? Howard. -- Automatic Disclaimer: The views expressed above are those of the author alone and may not represent the views of the IBM PC User Group. -- hdrw@ibmpcug.Co.UK Howard Winter 0W21' 51N43'