I tried mounting filesystems with -o noatime and that didn't help. I
then ran update (bdflush) with -f 300 to set the timeout to 5
minutes. That worked. It still seems odd that bdflush (1, NULL) will
write to the disc if *no process has written to the disc since the
last call to bdflush*. Doesn't the kernel properly tag all buffers as
clean once bdflush is done?
On another note, I set my HD spindown time to 30 seconds on my main
machine (not a laptop) earlier today, and late tonight the drive
stopped responding! I got drive not ready errors, and when I rebooted
a few times, things got progressively worse, first making unpleasant
noises and now the drive ID string appears to be corrupted (Linux
prints some characters which would be fine for playing cards, but not
for properly identifying the drive). The BIOS doesn't even see it at
all! I could possibly imagine that spining a drive up and down could
cause it to fail to spin up properly, possibly even giving media
errors, but why the drive ID string (which is presumably in firmware)
would be corrupted is a mystery. Has anyone ever seen anything quite
like this? The drive is a Maxtor 7425AV (420 MByte IDE), purchassed
about 2.5 years ago.
Regards,
Richard....