Can some hardware guru out there give an educated opinion about whether it's better to power down the drives to save energy/wear and tear, versus causing wear and tear because the drives are constantly starting up from a dead stop/powering back down again?

It's very much a horses for courses thing. If you are running some sort of backup-to-disk system, you are unlikely to have much activity on most of the disks most of the time, so you will definitely save money. If, however, you have a big-assed database that is being interrogated all of the time, you are highly likely to loose out if you try to power it's disks down when not in use, because this won't be for very long at a time.

This is the sort of system that you really need to do the maths before implementing.

Great idea, been looking for something like this for ages as part of our pseudo-HSM. The difficulties in developing a solution must have been huge though! Hope they've worked them all out.

Reliability: Few enterprise arrays will stay in service long enough to see any issues with high rates of spin up/down or load/unload. The vendor will have qualified the disks for this application anyhow and this is an easy one to do an accelerated lifetime test on.

Power: Turning a disk off in and of itself is only a quarter of the story. The biggest impact is really on the need for cooling (and UPS runtime sizing).

Applications: I'm sure the system will profile usage over time and adaptively manage the power policies. The big issue would be the delay in IO completion while the array spins up - perhaps heavy caching is used to relieve the problems caused by high latency IO (effectively a blocking IOP in-thread).

Interesting to see further info on this one..

For the latency, I wonder if perhaps they considered just slowing the disks instead of completely stopping them - this would reduce the initial spin-up current required and also mean the disks are not cooled too much.

Who knows - perhaps the next generation will be solar powered as well!

The question of whether this is worth it depends on the cycle time, and is relatively simple to calculate.

Answer three questions:

1) How much energy is required to spin up the drive(s)?

2) How muich energy is used per second when is use?

3) How long is it going to stay off?

1 and 2 are dead easy, 3 is the tough one - by the time you know that you would have saved energy, it's too late to do so. So you have to predict it.

A few years ago I spent some time doing essentially the same calculations for a compressed air system - calculating the cost of pumping up the system from empty versus the cost of leaving it running over the weekend.

Surprisingly, it turned out to not really be worth it.

Same question, different context.

For myself personally, the energy costs in this really make no significant difference. Keeping a HDD alive and running and not having to swap it out is worth more than energy cost savings.

That's fine, until you have over a certain ammount of storage. I only really know EMC's DMX arrays, but you are looking at 3phase power, tens of amps at a time, you _really_ notice turning off a %age of a couple of hundred disks.

Where I work we have something like 18DMXs, that's a whole lodda power, and cooling. We mainly don't bother with disk based backups, because a tape library doesn't use a fraction of the power/heat footprint. We do however, keep two snapshots of each server, one on two different arrays. Now the trouble with DMX here is that each disk is chopped up into virtual devices, these are then presented to the different machines, this means that you have to make sure that all of the devices on each disk are presented to machines which won't be useing them at the same time. All you have to do is accidentally stick a database onto a disk containing snapshots and it'll never power down.