The Myth Of The Supersonic Disk Drive

Over the past few weeks, I’ve heard folks promoting solid-state storage solutions and repeating the myth that spinning disks haven’t gotten faster than 15K RPM because it would mean that the airflow over the heads would become supersonic. While it’s a good story, and the thought of little tiny sonic booms inside a disk drive is amusing, there’s very little truth to the story.

For anyone who remembers his or her high school math, it’s pretty simple to calculate the linear velocity of the outer edge of a spinning disk. The formula is pretty simple, where LV is the linear velocity and RV is the disk’s rotational velocity:

LV=D*Π*RV

Of course since nominal disk diameters are normally given in inches and rotational velocities in RPM, the formula above will give its results in inches per minute. To convert to miles per hour we divide by 63360, the number of inches in a mile, and multiply by 60, which I assume you remember is the number of minutes in an hour.

For a 3.5” 15K RPM disk, this works out to just under 160 MPH. Since Mach 1 -- the speed of sound at standard temperature and pressure -- is roughly 760 miles per hour, that’s just 0.21 Mach, well below the roughly .8 Mach threshold where transonic effects raise their ugly heads.

It’s no coincidence that platters shrunk as disk drives developed from IBM’s original RAMAC, which used 24” platters, to today’s 2.5” devices. The last SLED (Single Large Expensive Disk), IBM’s 3390, spun its 14” platters at just 4200 RPM, a rate we now consider slow even for a laptop drive. Those big platters took a lot of power to spin, and a 14” 15K RPM drive would be doing .83 Mach at its outer edge, where transonic effects could cause problems.

If Western Digital had brought the 20K RPM 2.5” disk it was rumored to have been developing a few years ago, the linear edge speed would be just 120MPH, as the smaller diameter more than offsets the higher rotational velocity.

At this point, I must note that when this topic came up on Twitter a few weeks ago, Alex McDonald of NetApp pointed out that the air flowing around the drive’s head isn’t moving at the speed of the disk. I accept that, and accept that the computational fluid dynamics for calculating just how fast the air is flowing are beyond me. That said, even a 30,000 RPM 2.5” disk would be moving air just slightly faster than today’s 15K RPM 3.5” disks, making transonic effects highly unlikely.

[Read why Howard Marks thinks spinning disks will still be the better bargain through 2020 in "SSDs Cheaper Than Hard Drives? Not In this Decade."]

If supersonic heads don’t prevent disk vendors from spinning disks faster why, 10-plus years since the first 15K RPM disks, don’t we have faster disks?

Certainly the emergence of flash memory based storage systems and SSDs have made development of faster disk drives a fool’s errand. Even a 50,000 RPM disk would be much slower than an SSD. Now that system designers can use flash to absorb all the random I/O, there just isn’t any demand for faster disks.

The SSD effect explains why no disk vendor would spend the millions of R&D dollars developing a faster drive, but drive performance stalled long before flash became a reasonable alternative. What’s the real problem with faster disks?

That problem is power. The power required to spin any given disk faster increases at the square of the rotational velocity. So while a 30,000 RPM disk would have half the rotational latency than a 15K disk, it would take four times the power to spin the disk. Since spindle motor power is generally 75% or more of total disk drive power, the drive would draw at least three times as much.

Even then the performance benefit would be minimal. Sure, doubling the rotation speed will cut rotational latency from 2ms to 1ms. However random I/Os not only wait for the right sector to come around under the heads, but also move those heads to a different track. On Seagate’s fastest disks, the head motion takes an average of 2.6ms, so total random I/O latency on a 30K RPM drive would be 3.6ms compared to 4.6ms for the 15K drive, or just 22% faster.

So, no children, you don’t have to be afraid of supersonic disk drives. The SSDs made them irrelevant.

[Find out how flash-based SSDs work and the various ways they can be deployed in Howard Marks' workshop, "Deploying SSDs In the Data Center" at Interop Las Vegas March 31-April 4. Register today!]