Energy efficiency is big news these days, and everyone wants to jump on the bandwagon. We have a Green Storage Initiative within the SNIA, there’s The Green Grid consortium, and one of these days we may even have useful metrics on which we can compare different products. What’s it all worth?
A few months ago “green storage” was pitched as a way to feel good about yourself, in a CSR sort of way. Today, however, reducing operational costs is more important than ever, and something to consider before making additional storage hardware purchases.
On top of this, most of the large businesses I’ve spoken with in the past year have serious data center capacity problems. Not floor space — there’s more than enough of that to go around. The problem is power. They can’t add a rack of equipment unless they can get rid of at least a rack at the same time, because as compute nodes have become denser and more power hungry they’ve simply run out of spare amps on their available circuits.
So, suddenly everything’s turned green. And everyone’s turning green — storage vendors with envy at their competitors’ press releases, and pundits from getting sick of the press releases. Over at Drunken Data, Jon Toigo’s fed up with trying to save the world through disk, and LeRoy Budnik is tired of technology evolution being considered green. Why should moving to terabyte drives be considered a green move? You were probably going to ship them anyway…
Normally I agree with LeRoy, but I think drive density is in many ways a special case simply because of the magnitude of the change.
The Meaning of Drive Improvements
It’s usually a bit disingenuous to consider evolutionary advances as significant improvements in efficiency. For example, GE has a huge marketing initiative called Ecomagination, touting their commitment to the environment (and lower costs). Two years ago they introduced their new Evolution locomotive, hailing it as a major product of this green initiative. How much more efficient is it? Three percent. But what should you expect from a company pushing “clean coal”? (Guess it only has to be 3% cleaner.)
Hard drives, on the other hand, have been increasing in density at around 100 percent every 18 to 24 months. The power consumption for these larger drives doesn’t go up, so you’re essentially halving the power consumption per gigabyte, considering the drives only. That’s a huge jump.
For comparison, the 1 TB explicitly “green” drive from Western Digital consumes 7.4W operating. It’s non-green sibling? 8.4W operating. 12% is nothing to sneeze at, but it’s not as significant as 100%, and seek times suffer with these lower RPM drives.
With mechanical drives, there will always only be evolutionary improvements. It turns out that capacity increase has been the biggest win here! It’s been much easier to double the capacity of a single drive, thus halving the power consumption per GB, than halving the power consumption of a drive at the same capacity point. This will likely continue to be the case for a few years more.
On top of this, consider all the things that go around those drives; the controllers, the chassis, the RAM, the I/O interfaces. A modern processor consumes as much as 80W. FB-DIMM memory consumes 8W. A single 500 GB drive may only consume 8W of power, but the “fully-loaded” consumption for just that drive, considering it’s share of the rest of the storage system, may be as high as 30W or more. Doubling the drive capacity cuts this in half, while just halving the drive power consumption would only save 4W, or 13%. This is also why drive spindown isn’t a panacea — even spinning down the drive only saves 27%.
The other reason that hard drive capacity advances should be considered more than just evolutionary is because it’s actually very hard to make use of larger drives in conventional storage systems. I’ve already beat the RAID rebuild drum to death, but there’s a significant challenge just for normal I/O. While drive capacities have been increasing exponentially, seek times have not significantly improved. Without innovative file system layouts, it becomes impossible to efficiently make use of the available I/O bandwidth to the drive. I’ll comment more on this in the future.
The Next Green Frontier
If and when we run out of technology for shrinking our magnetic domains, the next frontier will be new storage technologies. Solid state disks have the opportunity for great additional savings due to near-zero idle power draw, and there are many likely paths forward for reducing their active power consumption beyond the levels already far lower than disk. At current rates, within two years SSDs will be cheaper than equivalent 15K RPM drives while being faster and more efficient. 10K drives are next on the block. Ultra-capacity drives will live the longest. Spinning disk is slowly reaching the end of its road, but there are many other paths that lead on.
In the mean time, if you’re unhappy with the rate at which disk power consumption is falling your other option is to reduce the amount of spinning disk needed in the first place. You probably can’t reduce the amount of data that you need to keep around, which is why we see deduplication as an inherently green technology, reducing the amount of disk necessary to store that same data. Deduplication saves power on the same scale as increasing drive capacity — 2X space savings is equivalent to a doubling of disk capacity. 10X space savings is equivalent to a 10X increase in capacity! This, in addition to an architecture that can properly utilize multi-terabyte drives, allows Permabit to provide low operational costs without the reliability risks associated with technologies like drive spindown.