Energy Efficient Storage or Energy Efficient Disks
by Hu Yoshida on Jul 24, 2007
One of the charts in EMC’s recent DMX 4 announcement was entitled, “Energy Efficient Storage Design”. This chart showed the kWh/yr energy consumed by different types of disk drives for a total capacity of 750GB.
Disk drive type kWh/yr/750GB
15K 73GB 5
15K 146GB 2.5
10K 300GB 1.2
7.2K 500GB 0.65
7.2K 750GB 0.43
The first number under disk drive type is the thousands of revolutions per minute and is a measure of performance. The second is the raw capacity of the disk drive in Gigabytes. kWh/yr/750GB means the total kWatt hours consumed in a year for the total capacity of 750GB.
Through the addition of the 750GB SATA disk they can claim a 91% improvement over the use of a 73GB FC disk drives for equivalent total capacity of 750GB. Basically this is comparing the energy consumption of 10.3 disks of 73 GB of capacity to one 750GB capacity disk. This energy saving is more a factor of the disk design than the DMX 4 storage system. While the ratios seem to be in order, I was surprised by the absolute numbers. 0.43 kWh/yr divided by 8760 hr/yr is 0.049 watts which is very, very, low. Disk drives usually run at about 10 watts when they are idling and about 15 to 20 watts when they are active doing read/write/seeking. That is pretty efficient when you compare that to a light bulb that uses 100 watts.
Consolidating 11 disks to one disk for the same amount of capacity will certainly save energy. But how much are you saving and what are you giving up? Can one disk drive do the work of 10 or 5 or even two disk drives when it comes to the I/O workload load?
While disk technology has been doubling capacity almost every year the performance characteristics have changed very little over the past 50 years. In recent years the technology advances in disk density have been slowing down so in order to continue the increase in disk capacity, the disk vendors have slowed down the disks and placed more tracks on the recording band. The larger capacity, slower, 7.2K disk has twice the rotational latency of a 15K disk. This rotational latency is experienced by every I/O to disk.
So if you need performance you might want to pay a little more to buy a few more drives of the faster, lower capacity disks and save energy in another way like turning off one of the light bulbs in the data center.
Another way to save energy is to use storage virtualization to non disruptively move less active data to lower cost, higher capacity disks and Dynamic (thin) provisioning to reduce the waste of allocated but unused space. Dynamic provisioning also enables you to spread your data evenly across hundreds of disks in a storage pool so that you can get more disk drives working together to service I/O requests. Improving storage utilization not only saves energy, it simplifies management and improves operational costs. This is an example of an Energy Efficient Storage design.
Comments (6 )
John- hardware vs. software RAID, RAID 5 or 10?…
Hu Yoshida touches on the larger number of small drives raid versus the smaller number of large drives issue in his post about power consumption here
The third column of data is actually the annualized Kilowatts per Gigabyte for each drive.
So if you multiply the 750GB drive number times 750, you get an annual power requirement of 324.12KWh. Divide that by 8760, and you find that the 750GB drive draws a rated 37 W/h (adjusting for rounding errors).
You can apply the same math to the other drives, resulting in drive power ratings which should match the spec sheets for the drives (top to bottom): 42, 42, 41, 37 and 37 watts/drive.
The calculations don’t consider idle vs. utilized power draw of the drives, nor the power demands of the storage array/controller, UPS/battery backup, cooling, etc.
Your assertion that wide striping of thin devices can mitigate the performance bottleneck contradicts logic: taking advantage of Dynamic Provisioning to double my utilization has the unavoidable inverse effect of also doubling the amount of work each drive has to do. It seems that you can either wide stripe for performance (at the cost of utilization), or you can condense onto fewer drives to improve utilization (at the cost of performance).
I’d also be interested to see the math behind your assertion that migrating low-utilized data off to external storage is more cost effective than migrating it onto other drives within a single array. Your external approach requires additional cache, logic, ports and power supplies on both the USP/USP-V and on the external storage, not to mention any hardware required to interconnect the separate storage devices. The drives are the same in both cases, the power/cost difference is in amount of incremental infrastructure required to support them.
Respectfully, I seem to have muffed my explaination of the watts/GB numbers – I thought I had it figured out, but the numbers just don’t foot to the drive specs.
Upon further inquiry, I’ve learned that at least some of the external & infrastructure power costs are included in these annual kWh/GB calculations, and that they are in fact measured power demand under load. Most significantly, they include a multiplier to account for the cooling (A/C) and overheads (losses) in the data center power distribution system (commonly accepted “shorthand” is that roughly half the power used in a data center is used to cool the equipment).
Hopefully, this clarifies the numbers – the intent was to show relative operational cost differences between the drive capacities in a given system. I don’t think you could use these numbers to do an apples-to-apples comparison of power requirements for these drives across different systems, at least, not without resolving all the unknowns of what the numbers include.
That’s an interesting chart, but I’m puzzled as to why they chose such an unusual denominator. I recognize that 750GB disks are an increasingly common size but wouldn’t it have made more sense to have normalized on something like 1 GB or 1 TB? It would be so much easier for a storage manager to then extrapolate energy usage and/or savings. It’s not like I can’t do the math, but why make me need to?
I find the discussion of drive-substitution-as-the-path-to-green-IT embedded in EMC’s announcement to be more than a bit of a stretch. So are rants from other vendors about virtualization, thin provisioning, de duplication and the like. These are all focused on throwing hardware at the problem. The only green hardware I have ever seen is the DELETE key.
I welcome you to a site I just built, launching this week in fact, called the Green Data Project. Green IT begins with Green Data. I think you will appreciate it.