PE> So if you have 8 data bits and 1 parity bit, then what you do is, say
 PE> your data is: 
 PE> 1 0 1 1 1 0 0 1

 PE> Well you then have a 9th bit (the parity bit) which records whether
 PE> the data has an even number of "1's".  In this case,
there are 5 1's
 PE> and 3 0's, which is odd, so you put in the parity bit a "0" meaning
 PE> "not even". 
 PE> Now when the cosmic ray zaps your data, and you have:

 PE> 0 0 1 1 1 0 0 1

  Yair, that's what I thought... it works on averages. If it's got one
of your 8-bits wrong there is a 1/8 chance that it got your parity
wrong, 1/64 that it got *two* bits wrong, etc, for a total of roughly
(I can't be bothered working it out) 1/6 or so overall.

  It only works if the RAM is *very* reliable, in which case why
bother? That's all I was saying: 1/8 parity checking only works if 
it hardly ever fails.

 PE> When one disk COMPLETELY DIES, you can tell what data used to
 PE> be stored on it, ie whether it was a 0 or a 1 (whichever one
 PE> makes your 5th disk's maths work out correctly).

  Yair... there's no magic in that when the data bit is a choice of
two and you trust however-many-discs to be true data.

  The problem with parity RAM is that we assume one bit has failed but 
none of the others did. What made that one-bit fail? Maybe they all 
failed.

Regards,
Bob
 
___ Blue Wave/QWK v2.12
@EOT:

---