Hard drive noises

[D. Hugh Redelmeier]

| From: Robert Brockway <robert-5LEc/6Zm6xCUd8a0hrldnti2O/JbrIOy at public.gmane.org>

| > still had the problem before applying a fix myself), and that my OS
| > endangers my hardware (even in what they refer to as "a very small
| > number of cases") disgusts me.
| 
| It may simply be an oversight, albeit a bad one.  Or they may have found
| enough problems with tweaking SMART settings that it made sense to avoid
| having it as the default.  Over the years Linux has had a number of apparently
| suboptimal defaults that were there because the optimum strategy was found to
| hard lock 1% of boxes on install.  The argument is that the installer should
| maximise chances of a successful install.

I don't remember exactly what the problem is.  But I'll talk as if I
do :-)

I think that Linux periodically accesses the drive (for sync?  but
why, if there is no change (assuming noatime)). 

These drives go to sleep (unload the heads) after a period
of inactivity.  This reduces power and heat and wear: all good things.
The length of this period can be set but has a default.  On some
drives, this default period is shorter than the Linux activity cycle
period.

So: in the normal course, the drive sleeps and wakes at a fairly rapid
rate.

Why did this start happening?  Because some drive manufacturers set
the default period to a very small number of seconds.  This isn't a Linux change
AFAIK.  I presume that this small number fits MS Windows behaviour.

So: who is to blame?  Disk drive manufacturers (if anyone).  At the
very least, they ought to have warned consumers or Linux distros or
someone.

So: what is the expedient fix?  Turn off the sleeping behaviour.  That's what
Giles did.

The correct fix (that someone might have done) would be to find out if
a certain threshold meshes with Linux behaviour and have Linux set the
drive appropriately.  Unfortunately, the meanings of the parameter
values are undocumented AFAIK and possibly different for each drive
model or drive manufacturer.

Perhaps the Linux periodic access is pointless and could be
eliminated.  That's the reason for noatime, but there might be other
periodic accesses that should be ditched.  I remember a periodic linux
access to CD burners that screwed some burners up (periodic query: "do
I have a CD loaded?").
--
The Toronto Linux Users Group.      Meetings: http://gtalug.org/
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://gtalug.org/wiki/Mailing_lists