> The following are excerpts from a bad/dangerous article about PC memory, and
> various Famous People at Microsoft commenting on said article. Names changed
> to protect... well, to be honest, to protect me.
[..]
> BL writes:
> If the mfg puts in parity, he gets blamed for
> the detected errors. If he doesn't, Microsoft gets blamed for
> failures caused by undetected errors. The incentives are all wrong.
Macs never utilised parity, bar a special-issue IIci for US government
orders, AFAIK.
IMO, decent error coding is sadly lacking from today's computers. Very
few of them will work reliably in orbit without the atmosphere as a
blanket to stop all those alpha particles causing single-event upsets
in memory, and in LEO we turn off the on-chip caches on transputers
etc and fly 386s for this reason. Error-correction done outside the
actual RAM chip really slows down performance.
Parity isn't correction - you'd need an extra bit to correct a
single-bit error in a byte - therefore it's completely useless.
There's no point in halting on detecting a parity error; you may as
well continue and hope the error was non-critical or even in the
parity bit itself.
Correction, not detection, is the key. Parity? Ignore it and hope that
the people implementing x8 chips actually have implemented some
correction onboard in the DRAM columns. (Have they?)
L.
<L.Wood@surrey.ac.uk>PGP<http://www.sat-net.com/L.Wood/>+44-1483-300800x3641