Re: memory & filesystem corruption under heavy load?

Robert L Krawitz (
Fri, 5 Apr 1996 10:18:48 -0500

Date: Fri, 05 Apr 1996 01:07:57 -0500
From: garth zenie <>

i can repeatedly, without fail, achieve filesystem corruption under
1.3.72-82 by unpacking two previously gunzip'd linux kernel source
trees while cat'ing eight copies of /dev/mem into seperate files
(probably don't even need to do all of that, it seems to happen by
itself just by unpacking a single source tree). running a diff over the
source trees reveals massive file damage.

What does the corruption look like?

One possible data point: when I use my fast memcpy routine I get more
corruption (this corruption does NOT look at all like the corruption I
got with buggy versions of memcpy -- that gave me blocks of null's
that were close to page boundaries, the current corruption is random
bits flipped here and there with no discernable alignment or other
pattern). I still have similar problems -- just less extensive --
without it. It's a somewhat rough data point because of the memcpy
routine, but it may be a data point nonetheless.

Also, I find that some days I get much more extensive corruption than
others (one day I had three files with corruption, another day I had a
few dozen). The difference seems to be more than could be accounted
for strictly by chance, and if I repeat the test immediately I have
the same thing happen. I think that I have more problems on very dry
days than on days with higher humidity, but I'm not positive. This
would tend to indicate a hardware problem, but it's interesting that
other people are seeing similar symptoms.

Is there anything in common here? Does everyone who's seeing problems
have, say, a Pentium, or do people with 486 and other systems see this
also? Are these older Pentiums (like mine -- my system is almost 2
years old)? Perhaps there's some sort of latent hardware problem that
only manifests itself under heavy load, and perhaps the improved
performance of more recent kernels triggers this somehow.

Robert Krawitz <> 

Member of the League for Programming Freedom -- mail Tall Clubs International -- or 1-800-521-2512