On Aug 13 2004, 11:27, Dwight K. Elvey wrote:
> You should try to create a more exhaustive RAM test.
> Although, time consuming, GALPAT is just about the
> most intensive.
I know Dave doesn't need it now, but a GALPAT is a very slow test for
any sizeable amount of RAM. It's also very old, and not 100% effective
against certain errors, though much better than some of the common
march tests. It was common in the 70s but a fair amount of work on
testing has been done since then and there are much better tests now.
I had to do a project for a degree course a few years ago, and found a
much better test (in Communications of the ACM), which I coded for a
Z80. It tests 8KB RAM in less than 6 seconds at 4MHz, but one of its
nice features is that it runs in linear time, so 32K would only take 4
times as long (24s) not 16 times as long -- a GALPAT has order O(n^2).
During the project, one of my colleagues worked out that the GALPAT
he'd coded would take an hour or two for the same memory. It also
doesn't need to use any RAM itself on a Z80 (because there are enough
registers), though it would on a 6502. If anyone wants a copy I can
give you the Z80 code, some notes from the project writeup, and the
CACM references.
BTW, most modern tests used commercially are less thorough than a
GALPAT; they generally depend on the assumption that most faults depend
on the bit(s) immediately adjacent to the faulty one, and require
knowledge of the cell topography to be properly effective.
--
Pete Peter Turnbull
Network Manager
University of York
Received on Sat Aug 14 2004 - 06:42:05 BST