28

I just assembled a new Linux file server with several Samsung HD103UJ 1TB hard drives and I am currently optimizing it.

The question I can't find an answer to is: is it reasonable to set a spindown timeout (with hdparm -S) for the drives? Can the spindown/spinup be harmful to the disk in the long run?

My previous file server did not have this timeout and the hard drives have been operational for 3+ years with no problems (always on) so I am not sure if I need this optimization at all.

What are your thoughts on this?

Thanks!

Peter Krumins
  • 3,435
  • 4
  • 21
  • 18

6 Answers6

16

I'm by no means expert on disk drive physics, but I see disproportionally many hard drives fail soon after the spinup. Probably the temperature change just does not serve them well. I would say: let them spin forever.

kubanczyk
  • 13,502
  • 5
  • 40
  • 55
  • 14
    +1 on spin forever. We once had to shut our server room down for the air conditioner to be repaired (it was scheduled maintenance). The EMC tech we contacted told us to leave the SAN running. He said we would have fewer drives fail due to heat than if we shut down and restarted once the work was completed. – KevinH Jun 21 '09 at 12:17
  • 3
    I've heard horror stories about long running SANs that had several drives with "stickage", after they powered it off then on again during maintenance. They ended up using a small sledge hammer to get them working again. *shudders* – Joseph Kern Jun 22 '09 at 13:05
  • A previous company that I worked for had a slew of HP NAS boxen (which happened to be the corporate standard for storage in the branch offices) die after shutting them down for any reason. Someone eventually established a policy of never allowing those Nas's to be shutdown... they were simply guaranteed to die if you did so. – Sean Earp Jun 24 '09 at 07:00
  • 3
    *> I've heard horror stories about long running SANs that had several drives with "stickage", after they powered it off then on again during maintenance.*   That’s probably because the drive was running for so long, the heat build up made it nice and soft, so when it turned off, the now soft materials stuck together. If the drive was kept cool enough, the metals, plastics, etc. would have remained hard.     *> Spin forever*   Data centers are probably one of the worst wasters of electricity today. – Synetech Jun 28 '12 at 03:22
  • 1
    "Sticktion" is a known problem from the old days - when powering down, the heads would stick to the platters, preventing it from spinning up again. Shaking the drive can frequently provide enough force to un-stick the heads, allowing it to spin up again. Modern drives don't suffer from this problem - if the firmware detects stuck heads, it uses its own motors (spindle and/or head positioning) to provide the extra force to get them un-stuck. – David C. Jan 06 '19 at 20:28
7

Is there any hard data on this phenomenon?

Frankly, this issue reminds me of my grandmother flipping out about using the phone in a lightning storm.

You always hear anecdotal accounts about how you need to keep drives spinning, but I've never seen anything based on verifiable data. I've run into a few instances where drives didn't come back after being powered down, but I've also run into many instances where everything came up just peachy.

duffbeer703
  • 20,077
  • 4
  • 30
  • 39
7

** Attribute #4 of S.M.A.R.T. is Start/Stop Count, which seems to indicate that start/stop count DOES play a role in disk fitness and when to expect a failure.

bobobobo
  • 769
  • 6
  • 14
  • 26
2

It's going to depend on the workload. You probably only want to spin the drive down, if it's not likely to be used for quite a while...

mibus
  • 816
  • 4
  • 5
2

I'm currently looking at the Power on hours, power cycles and Start/Stop counts for my HDDs. The drive in my desktop is set to power down whenever possible (since bought 5/6 years ago) and has had a default setting of after one minute for about 2 years. In that 2 years, start/stop counts approx 39,000 have overtaken the power on hours approx 12,000, and the the power cycles approx 11,000.

Currently I'm checking how many power cycles, start/stop counts and power on hours occur during an hour then a two hour period (in actual time). This will tell me how many times it powers on/off just browsing the net (not watching YT vids etc), and typing in MS Word etc - but as well as this, I wish to learn if a Power On hour means an hour for which power is supplied to the drive or an hour the drive is actually spinning. I will do another test with the computer in sleep/standby (but not hibernation/safesleep) in order to see if POH also increases when the computer is in sleep mode.

If you do something similar to me, you might be able to use this to decide whether to have your hdds power down. Download a SMART utility, one I really like in Windows is 'Crystal disk info' which is free, just bing it, and on Mac, SMART Utility, or SMARTReporter are the ones I have used. Crystal also allows you to get the smart status of external drives over USB (only disks not pen drives) - i found out that the old 6GB Quantum Fireball from and 1998 iMac had 8,900 hours and only a few hundred / 1000 power cycles. All my HDDS have been reported as healthy, apart from a 2007 Seagate laptop hdd i knew was dying.

I think its worthy of note that the I often hear the drive power down and up, sometimes the down is followed immediately by an up (but it always spins all the way down before spinning up).

original iPad
  • 21
  • 1
1

There even is a new trend in high-end storage : MAID (Massive Array of Idle Drives). Basically it's only big RAID arrays with support for spin down to save power... So if the big guns find it good for them, it may be good enough to you :)

wazoox
  • 6,782
  • 4
  • 30
  • 62