Re: files appear corrupt, but aren't; kernel bug?

[Dan Christensen <jdc@jhu.edu>]

Scott Henry <scotty@rahul.net> writes:
> >>>>> "D" == Dan Christensen <jdc@jhu.edu> writes:
> 
> D> Over the past few hours, my potato machine has been behaving very
> D> strangely.  Many of the files on my hard drive or which I read from
> D> cdrom have minor errors.  Specifically, at a random point in the file,
> D> two consecutive bytes are changed to 160 and 192 (240 300 octal).
> D> Moreover, this often repeats every 4194305 bytes!  (That 2 to the 22nd
> D> power, plus 1.)  Only large files seem to be affected, and this makes
> D> sense given the large gap between the bad bytes.  However, I believe
> D> that the files on the disks are fine because the problem goes away
> D> sometimes.  For example, during a period when the computer was
> 
> Hmmm, I would suspect either memory, or problems with the SCSI bus. 

Thanks for the suggestion.

It's definitely not a SCSI problem, because it happens on both my IDE
hard drive and on my SCSI cdrom.  And I'm doubtful that memory is the
problem, since the bad bytes are always 160 followed by 192, whereas
I would expect flakey memory to just flip a few random bits.  I
suspect that my memory is fake-parity but I don't know for sure.

Further info:  the problem seems to be completely gone now.  It lasted
for a few hours, and was consistent during those few hours.  Every
file I read that had bad bytes, consistently had the same bytes bad
until the problem went away.  Then the bytes were back to their
correct values.  I don't think this is a caching issue, since I
tested this on lots of very large files, larger than the ram I
have available.  And I have proof that I wasn't just imagining things:
while there were problems I copied a bad file to another file name.
Now the original is file, but the copy is still bad.

Very strange.  Does anyone have any other ideas?

Dan

-- 
Dan Christensen
jdc@math.jhu.edu