system excercising/burn-in tests?

Jim Kuzdrall gnhlug at intrel.com
Fri Aug 19 08:39:01 EDT 2005


On Thursday 18 August 2005 10:12 am, Bill McGonigle wrote:
> On Aug 17, 2005, at 16:14, Jim Kuzdrall wrote:
> > SuSE includes a memory test on their CD, but
> > Linux must have something someplace.
>
> I've heard that running a memory test under a booted system can only
> test X percentage of the memory because certain portions of memory
> are non-relocatable (kernel maybe?).  Anybody know if this is true? 
> I'm not sure if the distros put memtest86 on the disc for convenience
> or necessity.

    Yes, I agree.  In the old days, the memory test was built into ROM 
and operated from there, but now the ROM is unpacked to RAM.  On a 
modern OS, you never know where the system is taking its RAM memory 
from.  You can't go sticking test pattern bytes just anywhere, even if 
you replace the original data.

    Serious memory errors, like a faulty row or column driver will show 
up if you just get the biggest pool of memory malloc can give you with 
a minimal operating system.

    More subtle errors like a bit that follows the bit next to it would 
be very difficult to find.  One option it to use memory with error 
correction bits enabled.

    Fortunately, factories can tell if a memory array does not have 
enough performance margin from a statistical test of the chip before 
packaging, so "hard" RAM failures are uncommon.  Alpha particle 
discharge of the RAM cell (soft error) might steer a conservative buyer 
to less dense chips.

    I don't think I have observed a memory failure in years, although I 
see very few computers compared to most people on this list.




More information about the gnhlug-discuss mailing list