disk spin-up time

[Dong Li]

I got a question: is it possiable to make the disk spin-up time shorter than several seconds?

Currently, which kind of disks has shortest spin-up time?

Thank you very much!

Dong

[Dave Unverhau_1 · Accepted solution]

Dong,

Your question is pretty general, but I'll try to answer the question I think you're asking...

SCSI disks (in general) have a jumper to select a "delayed start". This allows disk arrays and JBOD chassis with several disks to stagger the startup to reduce power supply loading. The heaviest current load on a power supply occurs when the motor begins to spin the disk platters. If all drives in a disk array spin up at the same time, the power supply would need to provide a much greater amount of current to provide for this surge.

When the jumpers are set to "delayed start" (often labeled "dly" on the disk jumpers) the disk will wait for (10 seconds * SCSI ID) before spinning up. (The actual algorithm may vary between disk manufacturers...this one is coming from deep in my memory and may be very out-of-date.)

Anyway, if you're just using a couple of disks in a workstation or server and the power supply will not be taxed by the startup surge, you would want to be sure that "delayed start" is disabled.

Best Regards,

Dave

[Dong Li]

Dave,


Thank you so much for your reply.
I am designing an energy-efficient schedule policy for server side disks (array). We know conventional power management policies (saving energy by shutting down disks according to a dynimic threshold) are not workable on server side because the request interval (disk idle interval) is too short (e.g tens of ms) to shut-down and spin-up a disk without much performance degradation.


Though our I/O scheduling policy, we can stretch the disk idle interval to 3~5 seconds but it's hard to make it longer without degradating too much performance. While the spin-up time of the server disk is usually 10~20 seconds.

Can we find some disks with shorter spin-up time?

Thanks
Dong

[Dave Unverhau_1]

Dong,

I'm not sure that you would gain much benefit, if any, from what you propose, since the power required to spin a disk up is probably greater than what is required to keep it running for the few seconds of idle time most likely to be encountered in server environments.

You might want to do some power usage measurements of disks running steady-on versus those in a spin-up/spin-down cycle to see if further work on that kind of power management is worthwhile in a server environment.

Best Regards,

Dave

[Uwe Zessin]

Do you intend to power down the disk drive or send a STOP command? A disk drive that is powered up needs some time for self-diagnostics - that can include recalibration.

Honestly, I would forget this idea as I beleive you will damage the disk drives that way.

Today's high performance disk drives are amazingly aerodynamic wonders. The read/write head flys on a very small layer of air during operation. On a recent storage symposium I heard that the flight height is now down to about 10 nanometers! The diameter of a single atom is between 0.1 and 0.5 nanometers.

Even during normal operation the head can get partial contact to the platter. If you permanently request a spin down/up of the drive you cause the head to have intensive contact to the platter for extended periods of time.

A disk drive needs several seconds to spin down. In order to make the spin-down go faster it might take up the kinetic energy by turning the spindle motor into a generator and send the current through a resistor which turns the energy into heat. On the next spin-up you need to send valuable energy into the drive to re-build that kinetic energy - oops.