I looked up the spec for my HD - a Hitachi Travelstar 80GN -
and experimented. I'm using Fedora BTW.
First, the APM levels (set with 'hdparm -B') are grouped in
logical blocks: setting it to 128 or more (i.e., 80h) prevents
the disk ever going to standby mode (at least automatically);
similarly, setting it to 192 (C0h) or higher prevents the
mode ever dropping to low power idle - no matter how long the
period of inactivity.
The heads are unloaded in low power idle and lower modes.
So, for my drive at least, the assertion that, with a value of
254, the drive "will still unload heads, but far less often"
is not true. I would be suspicious of any other blanket
statements about this setting.
From experiments, it seems the raw value of load cycle for my
disk does indeed count transitions between low power idle and
active idle, i.e., head load/unload cycles.
However, the disk is still spinning in low power idle and
power consumption is 0.65W, not very much less than the 0.85W
of active idle, whereas recovery time deteriorates from 20ms
to 300ms.
Based on that, 'hdparm -B192' seems a reasonable setting -
higher might improve performance (at the cost of power) but
it can't possibly reduce the load cycles any further.
Load_Cycle_Count on my drive is almost at its threshold,
no other stat is anywhere near "old age" or "pre-fail".
As a raw number, it's over a million.
Further, it definitely looks like a hard drive problem,
not a Linux or even a BIOS one: my Hitachi's APM level is a
reasonable 128 at power up but anything lower
than 192 causes it to do quite frequent load/unloads - that
is, if it's allowed to drop to low power idle at all then it
will do so frequently and repeatedly without a prudent delay,
sometimes several times a minute.
The drive mode transition times are supposed to adapt to the
access pattern ("adaptive power save control"); I suspect
the algorithm is flawed perhaps because they haven't tested
against a sufficient number of usage profiles.
Nothing in the manual gives me any reason to believe the
drive's sub-192 behaviour is remotely healthy.
Slashdot Mirror
I looked up the spec for my HD - a Hitachi Travelstar 80GN - and experimented. I'm using Fedora BTW.
First, the APM levels (set with 'hdparm -B') are grouped in logical blocks: setting it to 128 or more (i.e., 80h) prevents the disk ever going to standby mode (at least automatically); similarly, setting it to 192 (C0h) or higher prevents the mode ever dropping to low power idle - no matter how long the period of inactivity. The heads are unloaded in low power idle and lower modes.
So, for my drive at least, the assertion that, with a value of 254, the drive "will still unload heads, but far less often" is not true. I would be suspicious of any other blanket statements about this setting.
From experiments, it seems the raw value of load cycle for my disk does indeed count transitions between low power idle and active idle, i.e., head load/unload cycles. However, the disk is still spinning in low power idle and power consumption is 0.65W, not very much less than the 0.85W of active idle, whereas recovery time deteriorates from 20ms to 300ms.
Based on that, 'hdparm -B192' seems a reasonable setting - higher might improve performance (at the cost of power) but it can't possibly reduce the load cycles any further.
Load_Cycle_Count on my drive is almost at its threshold, no other stat is anywhere near "old age" or "pre-fail". As a raw number, it's over a million.
Further, it definitely looks like a hard drive problem, not a Linux or even a BIOS one: my Hitachi's APM level is a reasonable 128 at power up but anything lower than 192 causes it to do quite frequent load/unloads - that is, if it's allowed to drop to low power idle at all then it will do so frequently and repeatedly without a prudent delay, sometimes several times a minute.
The drive mode transition times are supposed to adapt to the access pattern ("adaptive power save control"); I suspect the algorithm is flawed perhaps because they haven't tested against a sufficient number of usage profiles. Nothing in the manual gives me any reason to believe the drive's sub-192 behaviour is remotely healthy.
I'm going with 'hdparm -B192 -S60 /dev/hda'