On Wed, Jan 03, 2007 at 11:36:23PM +0200, Mihai Chelaru wrote:
> kefren64# md5 test.iso
> MD5 (test.iso) = 68345614422a119ad1510d3e570769c8
> kefren64# md5 test.iso
> MD5 (test.iso) = 93f0c39145f226a78c87ce4455242689
> kefren64# md5 test.iso
> MD5 (test.iso) = 4c83bbaa1d230c24642dddf5b9a4c9eb
> 
> I don't suspect any memory/HDD corruption regarding this particular server. 

Test it anyway.

I had the same behaviour on a raidframe install, FFS not LFS though.

During debug, I tried dd'ing a chunk off the start of a large
file to test, and repeatedly md5'd it. Once the chunk crossed the 700M
mark or so, I started getting the inconsistent results.

With older systems with little memory, I had 'learned' that memory
errors would likely show up through executables dying - but with new
machines with more than a gig of Ram, it's not true any more. You can
have an uptime of months and never see a SEGV, and still have bad RAM.

Now I memtest86+ every new machine and every new piece of RAM before
it goes into production.

-- 
David Maxwell, david@vex.net|david@maxwell.net -->
An organization gets what it rewards.
			      - Perry Metzger