HP Compaq 2710p notebook suffers too many head unloads (fwd)

D. Hugh Redelmeier hugh-pmF8o41NoarQT0dZR+AlfA at public.gmane.org
Fri Jan 11 07:01:54 UTC 2013


| From: David Mason <dmason-bqArmZWzea/GcjXNFnLQ/w at public.gmane.org>

| I have 2 WD Caviar Green, on which I run RAID1

These drives are not suitable for RAID.  The problem is that error
recovery done by the firmware takes an unbounded time.  RAID systems
take that as a sign of the whole drive failing whereas what you
actually want is the sector to fail and RAID deal with just the sector
failing.

If errors happen quickly enough, you don't have time to replace the
dropped drive and rebuild the array before the next one.  Once you've
had n such failures (normally n = 2), you are screwed: RAID has run
out of redundancy and has actually failed.

WD has a feature to fix this problem: TLER is the name (Time-Limited
Error Recovery).  They only support it for expensive drives.  Even
though it is dirt cheap to provide.  Welcome to price discrimination,
a sure sign of a monopoly market (the 3rd and 4th largest drive
manufacturers didn't limit this to expensive drives but they got
bought by the 1st and 2nd).

The best bet for RAID, AFAIK is the new WD "Red" drives.  Kind of like
Green but more expensive.

|, that I bought to be a
| matched pair.  But they have surprisingly different values on a few SMART
| parameters.  Turns out they're actually different models (which may not be
| a bad thing).

They are quite different.  For example EARS drives have (secretly) 4k
sectors.  Linux can handle 4K sectors, but only if the drive doesn't
hide the fact.  The main consequence is that you want to align
partitions on (secret) sector boundaries or performance is really bad.

Linux now handles this by having (some? most?) partitioning tools
round starting points to something like 1M (I don't remember the the
exact value).

Older tools often did the opposite: the partition sizes
were not a multiple of 4k bytes since they tried to align on cylinders
and cylinders were notionally an odd number of 512-byte sectors:
	255 tracks * 63 sectors/track
which is congruent to 1 (mod 8).  So only every eighth partition,
starting with the first (I think) would be aligned.

| sda doesn't have the problem that Hugh cites, but sdb appears too.... it
| looks like it's had over a million load cycles!!

Yikes!

| The
| https://ata.wiki.kernel.org/index.php/Known_issues#Drives_which_perform_frequent_head_unloads_under_Linuxpage
| suggests that hdparm won't let you do a -B, and sure enough, I get:
|           HDIO_DRIVE_CMD failed: Input/output error
|           APM_level = not supported

I don't actually see your drive listed in that table.  Perhaps you
should add it (I added a row for my system).

| I don't have any windows machines to run the wdidle3.exe tool that the web
| page suggests.  Any suggestions?

Try phoning WD support.  Is your drive in warranty?

Try googling: there are some sites like hddguru that expose a lot of
tricks and tips.  (I once worked hard to recover from a Seagate
firmware fiasco on many 7200.11 drives and found places like that
useful.)

Do report back!

I think that I have some EARS drives on mythtv boxes.  I had better
check them.

Thanks.
--
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





More information about the Legacy mailing list