Strange things going on with 4.8

[Robert Gray]

I'd like to emphasize that memtest86 doesn't catch lots of
memory problems.  Just last week I was having trouble compiling
mozilla so I ran memtest86 over night.  Nothing showed up.
But, "make buildworld" repeatedly failed on 
compiler signal 11 errors at about 20% complete.
Using  "make buildworld", I was able to isolate a 
bad DIMM and now "make buildworld" and
building mozilla run to completion (multiple times).

Whenever possible, I run with parity/ECC on the motherboard
and the memory modules.

I'm hoping a hardware/memory/motherboard expert will chime in.
How can manufacturers continue to make PCs without memory
checking?  With today's standards of 128-256MB in a PC, isn't
it just a matter of time until a bit gets flipped the wrong way?
Are manufacturers hoping that the bad bit will go unnoticed
in multi-media?  Is there something in today's
non-parity memory modules that helps insure reliable data?
Until I hear otherwise, I'll continue to spend extra
for the redundant, error-checking memories.

Thanks
-robert gray




Wes Peters <wes at softweyr.com> Sun, 10 Aug 2003 23:31:57 PDT says:
>>
>> Well the problem with testing memory with software is that its not
>> necessarily possible to hammer it hard enough to trigger the problem. 
>> If you can reproduce it easily you might try cycling out one dimm and
>> then trying to crash it. If removing a dimm fixes it then you probably
>> took out the bad one.
>
>In fact, many people in the FreeBSD community feel the best memory test of 
>all is to 'make world' several times.  I have experienced this myself 
>only once, but after returning the SIMM module to the vendor he verified 
>it was bad using a hardware tester.  The replacement SIMM has been in for 
>5 months now and the machine has been marvelously stable, as I expect 
>from FreeBSD.