Re: "green" drives ( was Re: Moving a PV onto raid)
On Wed, Apr 11, 2012 at 09:36:43AM +1000, Aníbal Monsalve Salazar wrote:
[ Cc-ing madduck as he may be interested on Craig's comments above ]
btw, i had already disabled idle spindown with hdparm -S 0 idle3-tools just updates the firmware settings so that it's disabled by default (as should be the recommend setting for WD Green drives on any linux system).. I had experimented with spindown over summer (didn't want my drives to overheat) but found that ZFS didn't like it. at all. kept getting drives being booted from the zpool seemingly at random, so ended up disabling it. of much less importance, I also found that hddtemp woke the drives up anyway, whenever it queried the temperature. [...] echo disabling disk idle spindown for disk in $DEVICES ; do /sbin/hdparm -q -S 0 "/dev/$disk" done [...]
It's a Debian package since yesterday evening.
nice :) craig -- craig sanders <cas@taz.net.au> BOFH excuse #314: You need to upgrade your VESA local bus to a MasterCard local bus.
On 11/04/12 11:17, Craig Sanders wrote:
On Wed, Apr 11, 2012 at 09:36:43AM +1000, Aníbal Monsalve Salazar wrote:
[ Cc-ing madduck as he may be interested on Craig's comments above ]
btw, i had already disabled idle spindown with hdparm -S 0
On the WD Green disks, all the smartctl APM-related commands are NOOPs. You can disable the spindown, or set it to a certain amount, but your drive wouldn't obey that at all. (That was my experience of fiddling with those settings, anyway)
idle3-tools just updates the firmware settings so that it's disabled by default (as should be the recommend setting for WD Green drives on any linux system)..
Not quite, see above. Am curious if you actually managed to adjust stuff -- were those all "green" branded disks? Which models? (Just wondering if WD has relented and let the OS adjust things now)
I had experimented with spindown over summer (didn't want my drives to overheat) but found that ZFS didn't like it. at all. kept getting drives being booted from the zpool seemingly at random, so ended up disabling it. of much less importance, I also found that hddtemp woke the drives up anyway, whenever it queried the temperature.
[...] echo disabling disk idle spindown for disk in $DEVICES ; do /sbin/hdparm -q -S 0 "/dev/$disk" done [...]
It's a Debian package since yesterday evening.
nice :)
craig
-- .signature
On Wed, Apr 11, 2012 at 11:50:33AM +1000, Toby Corkindale wrote:
On 11/04/12 11:17, Craig Sanders wrote:
btw, i had already disabled idle spindown with hdparm -S 0
On the WD Green disks, all the smartctl APM-related commands are NOOPs. You can disable the spindown, or set it to a certain amount, but your drive wouldn't obey that at all. (That was my experience of fiddling with those settings, anyway)
it's hard to tell for sure if/when the drive is lying to you, but setting 'hdparm -S 0' seemed to stop the mp2sas driver and ZFS complaining about drives being non-responsive craig -- craig sanders <cas@taz.net.au>
On Wed, 11 Apr 2012, Craig Sanders wrote:
On Wed, Apr 11, 2012 at 09:36:43AM +1000, Aníbal Monsalve Salazar wrote:
[ Cc-ing madduck as he may be interested on Craig's comments above ]
btw, i had already disabled idle spindown with hdparm -S 0
idle3-tools just updates the firmware settings so that it's disabled by default (as should be the recommend setting for WD Green drives on any linux system)..
I had experimented with spindown over summer (didn't want my drives to overheat) but found that ZFS didn't like it. at all. kept getting drives being booted from the zpool seemingly at random,
Er. How do we fix this? I was very much planning to implement zfsonlinux on my miniature NAS. That I want to keep operating in summer with good reliability despite my non-climate-controlled house. Mdadm and DM don't care about how long it takes to spin up a disk. This Is How It Should Be.
so ended up disabling it. of much less importance, I also found that hddtemp woke the drives up anyway, whenever it queried the temperature.
I have never encountered that problem. I do recall reading output that said paraphrased. "disabled, spun down". Perhaps it's a setting in the hddtempd daemon. Hey, look at that: fs# hdparm -y /dev/sda /dev/sda: issuing standby command fs# hdparm -y /dev/sdd /dev/sdd: issuing standby command fs# hddtemp /dev/sdc /dev/sdc: WDC WD10EAVS-00D7B1: 29°C fs# hddtemp /dev/sdd /dev/sdd: WDC WD10EAVS-32D7B1: drive is sleeping fs# hddtemp /dev/sda /dev/sda: ST9160310AS: drive is sleeping -- Tim Connors
On Wed, 11 Apr 2012, Tim Connors wrote:
On Wed, 11 Apr 2012, Craig Sanders wrote:
On Wed, Apr 11, 2012 at 09:36:43AM +1000, Aníbal Monsalve Salazar wrote:
[ Cc-ing madduck as he may be interested on Craig's comments above ]
btw, i had already disabled idle spindown with hdparm -S 0
idle3-tools just updates the firmware settings so that it's disabled by default (as should be the recommend setting for WD Green drives on any linux system)..
I had experimented with spindown over summer (didn't want my drives to overheat) but found that ZFS didn't like it. at all. kept getting drives being booted from the zpool seemingly at random,
Er. How do we fix this? I was very much planning to implement zfsonlinux on my miniature NAS. That I want to keep operating in summer with good reliability despite my non-climate-controlled house. Mdadm and DM don't care about how long it takes to spin up a disk. This Is How It Should Be.
For fuck's sake. Responses on the web seem to be:
I am wondering whether it might not simply be a timeout issue, that is: the drive is taking too long to spin up, which causes a timeout and a read error to be reported, which then disappears completely once the drive has spun up.
The power management settings put your drive to sleep after some time of inactivity.
Unfortunately the only way I have found to adjust this is from a windows pc utility. (You can download it from their website)". No, the disk isn't taking too long to spin up. The timeout is too short. How do we convince the ZFS code to be slightly more reliable? Hopefully it's in the manpage. I haven't yet progressed as far as working out how to install zfsonlinux, but if this happens to be impossible to configure, then there's no point in me looking any further. I *know* that mdadm and ext4 already do the right thing (I was just hoping for something "better"). -- Tim Connors
On Wed, Apr 11, 2012 at 03:34:32PM +1000, Tim Connors wrote:
I had experimented with spindown over summer (didn't want my drives to overheat) but found that ZFS didn't like it. at all. kept getting drives being booted from the zpool seemingly at random,
Er. How do we fix this? I was very much planning to implement zfsonlinux on my miniature NAS. That I want to keep operating in summer with good reliability despite my non-climate-controlled house. Mdadm and DM don't care about how long it takes to spin up a disk. This Is How It Should Be.
the issue is my LSI mp2sas controller card still being in RAID JBOD mode because I haven't been able to reflash it to IT mode in my current motherboard. This means it has drive seek & read timeouts tuned for enterprise drives in a data center that don't spin down except on explicit request, not consumer grade drives. anyway, the card times out and reports an error. ZFS sees the error and takes it at face value (as it should). as a i said, it's specific to my particular setup. it's probably fine on motherboard sata ports. and 'hdparm -S 0' did stop the problems so, from my POV - problem solved. also, fans. my hot swap bays have decent 120mm fans behind them, so i'm not too worried about them overheating even on 40+ degree days. i wouldn't expect any off-the-shelf mini-nas box to have decent cooling. that's one of the reasons i built my own. cost is the other major reason, i could build my own for a small fraction of the price of an OTS model and end up with a bigger, better, faster, more scalable system.
it. of much less importance, I also found that hddtemp woke the drives up anyway, whenever it queried the temperature.
I have never encountered that problem.
try idling a drive, then querying the temp with hddtemp or smartctl. It will spin it up. I haven't yet found a hard-disk brand that doesn't do this - so far tested seagate, western digital, hitachi, and samsung. craig -- craig sanders <cas@taz.net.au> BOFH excuse #350: paradigm shift...without a clutch
On 11/04/12 15:59, Craig Sanders wrote:
it. of much less importance, I also found that hddtemp woke the drives up anyway, whenever it queried the temperature.
I have never encountered that problem.
try idling a drive, then querying the temp with hddtemp or smartctl. It will spin it up. I haven't yet found a hard-disk brand that doesn't do this - so far tested seagate, western digital, hitachi, and samsung.
This might be something unique to your setup as well, Craig. My experience is in line with Tim's.. Querying the temperature responds with a message about the disk sleeping, and does not power it up.
On Wed, Apr 11, 2012 at 04:07:29PM +1000, Toby Corkindale wrote:
This might be something unique to your setup as well, Craig. My experience is in line with Tim's.. Querying the temperature responds with a message about the disk sleeping, and does not power it up.
you may both be right. and probably are. i don't recall right now the exact details of the problem with hddtemp and sleeping, just that they were incompatible. i'll have to look back over my notes but i had somehow remembered it as it spinning up the drive again. probably wrong. craig -- craig sanders <cas@taz.net.au> BOFH excuse #123: user to computer ratio too high.
On Wed, Apr 11, 2012 at 04:07:29PM +1000, Toby Corkindale wrote:
This might be something unique to your setup as well, Craig. My experience is in line with Tim's.. Querying the temperature responds with a message about the disk sleeping, and does not power it up.
you may both be right. and probably are. i don't recall right now the exact details of the problem with hddtemp and sleeping, just that they were incompatible.
i'll have to look back over my notes but i had somehow remembered it as it spinning up the drive again. probably wrong.
IIRC there was a bug in hddtemp around 2008 that it would wake up the drive. I'm not sure if that was really a bug or just assumed desired behaviour. Now I think you need to pass an explicit option to read the temperature of a sleeping drive (which will wake it up, making it warmer). Some people had a problem with some drives recording the hddtemp probe as 'activity' that would cause the drive to not go to sleep, but that's a slightly different problem, and I don't know if it was a bug in hddtemp or a bug in the drive firmware. James
participants (4)
-
Craig Sanders -
James Harper -
Tim Connors -
Toby Corkindale