[TAG] segmentation fault

[Benjamin A. Okopnik]

On Sun, Oct 02, 2005 at 04:26:13PM +0100, Martin J Hooper wrote:
> Karl-Heinz Herrmann wrote:
>  > I guess you could swap the two memory banks. That should put an error in
> >the memory chips in a different address location. It won't catch a messy
> >interaction of chip/socket. 
> 
> Won't memtest86 check it?? - I know it checks all the memory and that it 
> runs a series of tests.

Eh, maybe. As far as I'm concerned, it's nearly impossible to guarantee
that a software test will locate an intermittent hardware fault, unless
you run it within reasonable time parameters for that fault (i.e., if
you're averaging one failure per week, then you should expect to run a
two-week-long test as the worst-case scenario.) In most cases, that's
not a reasonable thing to do: two weeks worth of down-time for a
computer is usually worth far more than the cost of a new machine. Even
once a day failures become somewhat problematic when seen from that
perspective.

You can try the old technique of using a can of Freon and a hair dryer
(cooling and heating the SIMM alternately while running a memory test),
but, again, given the small number of possible points of failure (i.e.,
the number of SIMMs), the most efficient technique would be to swap them
out.


* Ben Okopnik * Editor-in-Chief, Linux Gazette * http://linuxgazette.net *