Contributing Author: David Floyer

Disk drives are the heart of storage systems. Over the years we have seen a migration of the form factor from 5.25" to 3.5" to the recent introduction of 2.5" technologies. So smaller disks mean less metal to move, less distance for actuators to move, and less air resistance. Do smaller drives matter for green storage?

First the “Yes” part -- Let's examine the physics involved. The equations below show the theoretical relationship between RPM, platter diameter and the power required to drive the platters:

• Power ∝ (#Platters) x (RPM)^2.8 x (Diameter)^4.6 [1] ("∝" means is proportional to)
• Power = constant x Diameter^4.6 x RPM^2.8 [2]

If we take the current actual platter size of the high-end drives as 3.5", and the new generation disk drive as 2.5", and keep the number of spindles and rotational speed the same, the reduction in power requirement is 40%. So, RPM’s and disk diameter make a significant contribution to the power required to drive a spindle -- roughly the cube and fifth power respectively. This does not include the power needed for the electronics or the actuator.

However, a smaller platter means less capacity for the same areal density; in this case the reduction in capacity is 31%. Consequently, overall, the effective reduction in power is 14% for the same capacity. Nice, but it isn't going to change the world.

Lets look at the space requirements. The space requirements for the 3.5" drives are 1" high, 4" wide and 5.75" deep [4]. The space requirements for the 2.5" drives are 0.6" high, 2.75" wide, and 4" high. Do the math and the overall areal density (width x depth) improvement is 50%. Twice as many disk drives can fit in the same space in a 1µ server or a laptop. Really good news for high-end PC and 1µ server vendors, who can attack the weakest link in system performance, disk performance and capacity. At the moment, 2.5" drives are more expensive, but at the high end there is a premium available for improved performance and capacity in a smaller footprint.

The next issue is performance. When the capacity is normalized, the performance of (say) two 3.5" form factor drives vs. three 2.5" form factor drives, the new drives deliver significantly more throughput from:

• 50% more actuators;
• 10% faster seek times (less distance to travel).

So can you have the same capacity on a 2.5" drive compared with a 3.5"? To get the same capacity per drive, vendors would have to add platters, and increase the drive height. Technically, this can be done, but this pushes up the power and cooling requirements. So much so in fact, that when one of these drives is placed in a modern 1µ server, the heat density generated exceeds the ability of the server to be cooled by air.

In disk arrays, however, the problem of cooling is not quite so acute as in 1µ servers. If you add in the height dimension as well for disk arrays and look at the volumetric density, and there can be three times the number of 2.5" form factor drives in an array. The heat density for 3.5" storage bays is less than 1kWatt per sq. ft. The maximum that can be cooled by traditional air cooling systems is about 10kWatts/sq. ft. So even a three-fold increase in heat density as a result of moving to 2.5" technology would not impact the ability to cool storage bays.

So in theory, there could be special models of 2.5" drives with extra platters just for array vendors. This is unlikely to happen because the volumes of devices going into arrays is relatively smaller than systems. In our discussions, we learned that the biggest volume demand for smaller drives comes from the vendors making PCs, 1µ and 2µ servers (not forgetting set-top boxes and other media storage), not from the disk array vendors. What’s worse, going to smaller form factors decreases the amount of heat that can be dissipated to the outside air. In addition, adding the additional platters would significantly impact the availability characteristics of the drive. Once again, this underscores the fact that air is a terrible coolant, and that high performance equipment in general is reaching the limits of current air cooling technology.

Figure 1 - WW Shipments of SFF Devices in Multi-user Applications

Industry forecasts for Small Form Factor (SFF) drives with SCSI/SAS and FC interfaces in multi-user applications show a strong shift to SFF over the next few years (See IDC's forecast in Figure 1)[5]. As 2.5" volumes increase, all research and investment in improving 3.5" technology should in theory dry up (although high capacity 3.5" SATA devices will have some life). If historic trends are a guide to how switching from one drive form factor to the next applies to this new generation, the shipments of 2.5" drives will exceed 3.5" drive shipments within 12-18 months for these high performance applications. At that point, 2.5" drives will improve and become lower cost than 3.5" drives for these applications. The drive manufacturers will be in a hurry to free up manufacturing space for the next generation of disk drives and will wind down 3.5" production as soon as they can. Again, the wild card remains high capacity SATA devices for less performance intensive applications.

If this scenario plays out, array vendors are likely to move to 2.5" drives within 12-18 months.

However, it is clear their capacity will not be equal to the bigger diameter drives. So, more, but smaller drives per array will:

• Reduce power by 30%/drive,
• Increase power density by 100% to 200% for an array,
• Continue the historic Power/GB curve the areal bit density on the platters increases,
• Increase the heat and cooling loadings by 100% to 200%,
• Increase performance and (maybe) enable the use of higher density drives.

What else can be done to reduce the heat output from drives? There has been much research and now a lot of talk about dynamic management – dynamic RPMs, dynamic thermal management etc. Dynamic RPMs were tried a few years back, but received poor market acceptance – as the drive got hotter, it slowed down programmatically. Turning off a drive that is not in use, a la MAID, is technically effective but extremely difficult to manage. Clearly turning off a spare drive in a RAID rank is effective, but a relatively small contribution. Other options exist and they come from features of laptop drives. Typical levels of savings that can be achieved are:

All the drive vendors have or are moving to add these kinds of features to their enterprise-class drives.

Another tactic is to use colder cooling air – if you can. In general, heat transfer is proportional to the difference in temperatures. It won’t save power, but it may improve sustainable packing densities.

Action Item: Storage managers should project their storage requirements over the next five years, assume a rapid transition to 2.5" drives, and calculate the power, space, power density and cooling requirements. They should then work with the data center designers to ensure that there is sufficient power and cooling available, and that there are the funds and commitment for the necessary infrastructure changes. In addition, storage managers should put in place a green storage infrastructure that will minimize disk drives by being able to:

• Turn off spares,
• Aggressively pursue tiered storage with lower RPM drives,
• Implement MAID and/or drive spin down where appropriate,
• Aggressively implement virtualization and thin provisioning to reduce the drives required,
• Consider rear door heat exchangers using chilled water,
• Consider the use of external storage services and/or outsourcing or storage equipment,
• Ask the vendors about their green storage commitment.

Footnotes: [1] “Power Management of Enterprise Storage Systems”, Sudhanva Gurumurthi, August 2005, PHD Thesis, The Pennsylvania State University. The Graduate School Department of Computer Science and Engineering
[2]”Technological impact of magnetic hard disk drives on storage systems”, E. Grochowski and R. D. Halem, IBM Systems Journal, VOL 42, NO 2, 2003
[3]”Quietly Cool”, A White Paper published by Hitachi © 2004 Hitachi Global Storage Technologies. Used by permission. Reuse of Hitachi copyright content is not authorized under the GNU General License posted in the Wikibon Terms of Service.
[4] Specification sheet for the Seagate Savvio 2.5" 10K drive

[5] IDC Worldwide Hard Disk Drive 2007–2011 Forecast Update