laser-lovers@cca.UUCP (10/16/85)
From: Chris Torek <chris@mimsy.umd.edu> Alas, testing every one out of every 1K bytes (or other large chunk) fails miserably as a memory diagnostic. I suspect that this is how Imagen's software counts up available memory in the first place; and I know for a fact that Unix counts up Vax memory using this method. Memory diagnostics are subtle things. There are `walking 1s', `walking 0s', `refresh delay', `address line short', `data line short', and other standard tests, yet these do not even begin to cover some of the interesting faults that can hide in modern dynamic RAMs. A good test program will try all of these and will also run a pseudo-random number generator. A truly thorough test takes on the order of eight hours per 64K! My information on dynamic RAM problems is dated, so my figures may well be off; but suffice it to say that a power-up self-test cannot perform a complete check. Perhaps what Imagen should provide is parity error detection and/or diagnostic floppies. Chris