Let’s Guess: Does Spindown Reduce Cost?

My granny always used to say; “One man’s meat is another man’s poison”. And Marc Farley’s Top 10 Network Storage Innovations clearly illustrated the saying when Tony Asaro stated that he had spindown on his list. 

“Spinning down drives - this makes it 11 but I do think this adds value and is innovative. We need to reduce power and cooling and this impacts that with backup and archived data.”

My list was the bottom ten storage non-innovations. At # 9, I had spindown.

“Another idea related to ILM that seems to have positive benefits; if, and only if, the cost of management is less than the cost of the power and cooling you save by spindown. I have yet to see a positive TCO case for this; micro-managing data is expensive. There are lots of other issues too, but that’s for another time.”

And at # 10 I had ILM, spindown’s big brother.

Let me explain that I’m not against spindown or ILM or anything else for that matter if it saves more than it costs to do.

Cries of “Well, obviously it must” don’t cut it with me. This is a major problem; how can we justify spindown, low energy this and that, green initiatives and a host of other blah-blah doo-dahs that we claim to reduce costs, if we can’t measure and model them?

I haven’t seen any costing models for ILM, or backup & recovery, for instance, nor have I found any on the intertubes. Perhaps I need to ask the right people. So help me here; anyone got any costing models they’d care to share that we can build spindown into?

Otherwise we all have to confess publically that we don’t know if any of this stuff can save power, cooling, money, the planet, and that all we can do is guess.

On that score, spindown and ILM stay in my bottom 10.

Comments and cost models welcome.

.

Comments

[Chuck Hollis]

Couldn't agree more. As a matter of fact, I'd extend that thought to data deduplication and other areas :-)

As far as modelling costs for spin-down, for some time now, EMC has offered a power calculator that's very popular with our customers. We already owned a vast array of EMC and non-EMC arrays, servers, switches, etc.

Rather than just go by spec sheets, we measured actual power draw at light, medium and heavy loads. We also did field measurements of what people pay for power, cooling, etc. since those numbers are not always on hand.

So, when we model cost-savings for spin-down, it's pretty easy to look at the delta between power draws, approximate power-on vs. power-off duty cycles, apply either the actual or estimated costs for power and cooling, and arrive at a pretty realistic number "here's what it's worth" without too much drama.

Not everyone can see cost savings -- true.

But many can, and being able to get to that number ahead of time -- using customer's actual data -- helps to justify the feature.

I think the tool has become most of our popular, simply because it uses actual measure power draw numbers, and covers all manner of data center equipment including EMC's. The power/cooling cost survey is pretty popular as well.

So, have you guys done anything to create a real-world power calculator that you can put in customers' hands?

Cheers!

[Alex McDonald]

Wow, that was quick. You have some form of hyperdrive keyboard attached to a time machine version of VMware on your laptop? Or are you reading my posts as I type them? Must check for spyware....

You make a good point about power calculators, but that's a fractional part of what I mean, and it highlights the problem.

Let's say your power calculator states that I'll save $3000/year by using spindown 23 hours out of 24.

Great! Let's do it.

But wait a moment; to utilise spindown, I need to have some form of data management strategy that lets me take advantage of it. I need to analyse my data & identify suitability for this treatment (cost), change processes (cost), move my data (cost), retrieve it when required (cost), transfer it to other media in the medium/long term as technology gets replaced (cost) etc etc.

For backup & archive, that's probably a little easier than most to assess, but only marginally, and it's only part of the lifecycle of data cost.

For first generation data right through to archive, it's not that easy -- or cheap to do. An example that springs to mind was your "suggestion" that NetApp should spindown some of its data based on research we'd done on access patterns, size and age of data we store in a CIFS environment. Apparently, we could save a shedload of money that way.

I don't think so. That's where a whole life cost model of data is needed, and an understanding of how it's used, frequency of access, cost of storage and so on.

Once we've done the analysis (cost), changed process (cost), moved data through tiers (cost), retrieved it when required (cost), transfered it to other media in the medium/long term as technology gets replaced (cost) and so on, our $3000/year saving may be not a saving at all.

What does it really cost to do something, as opposed to what I'd save by doing nothing? Having a power calculator doesn't answer that.

Many years ago in a galaxy far away I and some colleagues worked to develop a product to build ABC (Activity Based Costing) models for IT. I found some interesting models for digital document management in library management systems this afternoon that reminded me of them. I think I'll explore further.

As to dedupe, that's a part of the data management story. Right now, with disks still spinning beause no-one can take advantgage of spindown on primary, it's a big save, because it has such small marginal costs.

Unless, of course, it's not available. Or free :-)

[marc farley]

Alex and Chuck,

I agree that there aren't currently any big economic gains from spin down, but I think Tony was writing his list with a forward-looking perspective.

Spindown lacks a critical mass of software to make it work, but I believe that it will be developed (I don't know who will do it necessarily) to address the need corporations have to store the mountains of data that sporadically needs to be accessed - and is required by laws and regs.

Spin down works very well for me with several USB and Firewire disk drives. Just this week I happily re-started an audio project of mine that needed to come back from stationary dormancy. I like the fact that this data wasn't consuming any power these many months. For me, its removable, but in the corporate world I think spin down makes a lot more sense.

The same question could be asked about flash SSDs. I don't see that much of a requirement for them today either due to their current premium pricing and the economic conditions. Of course, their application is much easier to understand, - high performance storage. But disk drives have continued to surprise for many years and other less expensive technologies, such as wide striping could limit SSDs market penetration.

If the economic malaise drags on, I give spin down better odds than flash.

[Lee Razo]

Ultimately, the disks that consume the least power of all are the ones that never had to be used in the first place.

This has always been my thinking whenever COPAN would come up. Great that they can turn their disks on and off, but what have they actually done to reduce the amount of disks used in the first place?

These days it's looking more like COPAN itself has been spinning down.