Some folks who look at the speed difference between HDDs and SSDs ask themselves why HDD makers don’t defend their market against the SSD menace by cranking the speed beyond 15,000 RPM. An unfortunately popular answer is that the edge of the HDD would have to break the sound barrier and the HDD would shatter. Is this true?
With a little back-of-the-envelope calculation The SSD Guy can rapidly disprove this. (Forgive me, but I am going to mix metric and English measure here.)
The speed of sound is 340.29 meters/second. HDD diameters are measured in inches: 2.5″ and 3.5″ and speeds are in revolutions per minute, so let’s convert this to inches per minute. To do this multiply by 60 seconds/minute and multiply again by 39.37 inches per meter. That gives us 803,833 inches per minute.
A 3.5″ HDD has a circumference of 3.5 x π or 11 inches. That’s the distance traveled in one revolution. Divide this into 803,833 and you get 73,105 RPM. That’s pretty fast!
Naturally, when you go to a 2.5″ HDD the speed gets even faster. Its circumference is 7.85″, and 803,833/7.85 = 102,347 RPM.
So why don’t we see any 20,000 or 30,000 RPM HDDs on the market? I asked a number of HDD designers and got a very simple reply: It would take too much power.
As the speed of an HDD increases the power it takes to run it increases disproportionately. These designers told me that increasing the speed significantly beyond today’s 15K RPM level would cause undue headaches in power management and could even threaten the lifetime of the HDD.
Although the “speed of sound” argument is catchy, it’s just wrong. The power argument, however accurate, is very dull by comparison. But consider this – SSDs already trounce HDDs in measures of IOPS/Watt. Imagine a comparison with a 30K RPM HDD with perhaps three times the power consumption of a 15K model but only twice (or less) the IOPS performance. How would that compare in IOPS/Watt to an SSD? Th HDD would look far worse than even its 15K RPM counterpart!
Objective Analysis has published a report on enterprise SSDs: The Enterprise SSD: Technologies & Markets that can be ordered for immediate download from our website. This report discusses issues of IOPS/Watt and IOPS/$ and how these compare to enterprise HDDs.
Oddly enough, the photo accompanying this post is often said to be of an aircraft breaking the sound barrier. According to Snopes, this is also a myth, and the phenomenon really relates to a sudden pressure change that creates a condensation cloud surrounding the plane.
I don’t understand why they don’t use multiple arms on the same surface or use multiple stationary arms with a line of readers on it.
Hitachi did that years ago. The drives were very expensive (more than 2x a single drive with 2x the capacity) and the technology didn’t scale. The problem is that each head arm causes turbulence, especially when moving, and that makes flying really low (very high density) difficult to achieve.
There have also been various attempts at multi-head drum drives, which use a cylinder not a disk. These all failed due to cost.
Disk substrates are either very thin metal or glass. At 30K RPM, they would stretch and disintegrate. Higher speeds also magnify high-frequency flutter. The head just wouldn’t be stable.
Still, you have to marvel even at 15K RPM drives. They are reasonably durable, and before the advent of jet-propelled SSD, where the fastest flyers around. Now, it’s academic. There’s no way to catch up with SSD and flash. The only thing preventing a massive transition to flash is price (limited supply v. high demand). At 10x the price, SSD is still limited to the fastest storage portion of a storage pool. Will it change…..there’s a lot of sand around, but flash has other solid-state competitiion on the horizon that’s faster, denser and won’t wear out.
Bill, Thanks for the comment.
As far as I understand that’s a matter of cost. I would like to hear what others have to say about it, though, especially seasoned HDD designers.
I have heard that tolerances are so tight that one head can’t read tracks written by another, but you still could use two actuators (sliders, arms,…) if they covered different portions of the same HDD and that could double throughput.
As for Jim O’Reilly’s comments it’s clear that you know a whole lot more about this than I ever will! Thanks for providing great insights. I have to wonder, though: if there really were a market for a 30K RPM HDD wouldn’t the industry find a substrate that could withstand the high RPM? HDD makers and their suppliers strike me as very resourceful people.
I think the super fast hdd in the new age of mechanical hard drives should be titanium.
I am not a materials scientist, just speculating.
The Titanium disk may be able to withstand super high rpm beyond 15000. It may require a special coating
such as aluminum to register the electromagnetic “bits” imprinted on it by the read/write heads.
I’m sure somebody is already working on this. You Know: Government secret suckers and their buddies.
Dan,
Although the article argued that power was the reason that HDDs don’t rotate faster, you do bring up an interesting point.
So interesting, in fact, that I took it to the analyst who knows HDDs better than anyone: Ton Coughlin. http://TomCoughlin.com
He had this to say: “There is some merit to his ideas. A stiffer substrate will flutter less and thus can be made thinner or run at higher RPM. That is why many of the higher RPM HDDs use glass and glass/ceramic substrates (these are cheaper and I think stiffer than titanium). Less flutter results in less friction on the motor and thus less power consumption. However, I think the increase in RPM, if significant, would probably dominate the energy consumption of the drive.”
So energy is still the issue, but stiffness does help.
Thanks for the comment.
Jim
This has been very interesting. Thank you for the acknowledgement.
I would like to know which materials are stiffer than others and have less flutter if expense was not the prohibiting factor.
I would like to see the leading edge of the materials science that could be utilized in HDD development. Just because we can.
In some rare applications there might be a requirement for very low rotating mass/weight, near zero flutter, and
very high rpm tolerance.
Let’s depart from the domestic theatre, and also leave behind the collegiate level of science.
Consider for a moment if someone were pursuing the goal of building the next generation mechanical HDD for whatever esoteric reason it was. The goal is that the device shall have an undefined lifetime.
MIracle bearings made of diamond never wear out.
Imagine that all the problems with the read heads and the coils were solved with light.
Maybe the disk has a crystalline or chemical coating that reacts to laser light (0,1)=(red,green) on a spinning non-removable mechanical disk and reflects it back to a detector or series of detectors. Just speculating….
We’ve got the disk now spinning perhaps over 100,000 rpm and it is our new miracle material that does not shatter.
I’m hopeful for a new breakthrough in storage technology.
The reason I’m looking for this functionality is I think we need to send (“first contact”) a working greeting card out in space that won’t break before it gets to its destination, or somebody else finds it. We can’t shield the ssd from cosmic rays in the long term.
I have sometimes been disappointed with the Earthbound hardware I was using for projects.
I have bought one particular brand of 64 Gb usb SSD that failed to detect right out of the box, and the warranty replacement did the same thing. I do know how to use my windows disk management utility.
I am now avoiding that manufacturer for my projects.
The drives that I really want for my projects are beyond my budget at this time.
I’m happy to say that I’m fully committed to a Masters degree program in Information Technology
at UNLV this Fall.
“Luck is not a factor”
: )
Jim,
Do you have a link to the lab or group that is pursuing HDD research for the science of it, as opposed to consumer level devices? My previous comment was rather lengthy, but I would like to keep up with HDD research science.
Thank you,
Daniel
Daniel,
I took this to the guy who knows more about HDDs than any two people: Tom Coughlin of Coughlin Associates (www.CoughlinAssociates.com)
He listed CMU, Stanford, Univ of Alabama, Univ of Minnesota and York University. He added that there may be others in Japan and DSI in Singapore might possibly still be performing this kind of research.
I suspect that you could find the names of the research team leaders at these institutions without much effort.
Good luck!
Jim
Another point that you have not taken in to consideration is the capacity impact. Reading and writing to a disk is done charging a magnetic area on the disk as it rotates under the head. The amount of time required to ramp up and down the charging of the magnetic coil within the head determines the frequency of writes. The write frequency determines how much data can be written during a single rotation of the disk. As the disk rotates faster without changing the frequency of the Rd/Wr channel the bit density goes down, thus loosing capacity. Increasing the frequency of the Rd/Wr channel is also limited by the capabilities of the media.
If you do some simply math and calculate the time required to complete a single rotation of the disk. Looking at a drive that rotates at 7200 RPMs it takes 8.33mS per revolution (1/(7200/60)). If you take the same drive and rotate it at 10000 RPMs the amount of time drops to 5.99 mS. This is ~30% reduction in the amount of time that the drive has to write a single track. Without changing the capabilities of the head and media this is a ~30% reduction in the capacity. Each time you step the rotational speed of the disk up you either have to speed up the Rd/Wr channel and the capabilities of the media or you take a loss in capacity.
When accessing data in a HDD there are two sources of delay; Seek Time and Rotational latency. The rotational speed of the drive has minor effects on the seek time. Decreasing the rotational latency helps the head get to the requested location on the disk faster after completing the seek. Increasing the rotational speed only helps for small random IOs. For large sequential IOs this actually hurts the performance. The reduction in track capacity results in the drive having to reposition the heads more often. Each head or track switch takes a few mS to accomplish.
The HDD is a very under appreciated device for the level of complexity required to perform as it does today. The HDD is a very complex electro-mechanical device. There are simple answers to speeding the drive up, but each of these simple answers kick off a series of very complex mechanical and electrical interactions. Often the results have residual effects that cause other problems like increases in power consumption, heat, vibration or cost that exceed a level that the market will bear.
Hubert,
Thanks for a very solid and well-written reply. I had never considered how read channel bandwidth would be a limiting factor, but this makes perfect sense. Also your point about the impact of seek latency is very important. As rotational latency approached zero, it would do less and less to help, while the seek latency’s effect on overall performance would increase.
I shared a pointer to this post on LinkedIn and one reader (Chuck Still) provided yet another important consideration – storage array OEMs don’t want to add more power & cooling to their racks. Increasing HDD RPM would require this, so there’s some push-back from this side.
All very good inputs. Many thanks to all.
I was led to believe that 22K rpm drives were being sampled to Compaq in the 2000-2001 timeframe, so they obviously existed but have never made the market, i never did find out why, but the power issue if true would be the reason.
The point about HDDs these days is that they almost invariably form part a RAID array, have controller level cache for both read and write, and often drive level buffering too. Putting that together means the raw performance of a single HDD matters much less than it did. Doing a write mostly means putting the data in write cache and then signalling the application that the IO is complete, the actual writing out to the HDD is done in the background. Similarly for reads sophisticated controller algorithms will detect what type of read it is – random, sequential, long sequential, etc – and read ahead to stage data into the cache, again taking the HDD performance out of the critical path for IO. Spreading data blocks across larger RAID4/5/6 arrays also improves the performance an application sees from it’s data, a RAID group of 16 or more isn’t unusual so even in a dual parity RAID6 configuration the data is coming off 14 or more spindles.
Of course the performance of SSD can’t be denied. I’m old enough to have seen IBM 2305 data drums in use, with a read/write head for every track, but when StorageTek emulated the 2305 with an SSD box it blew the drum and successive generations of HDDs away. For paging on a mainframe it was unbeatable. If I was designing a storage infrastructure today I would certainly include SSD and the capability to intelligently pin datasets in in cache as part of the teired design. The only thing I haven’t done the arithmetic on is the total cost per GB of SSD against HDD, instinctively I assume the cost of housing and spinning 1TB on an HDD is going to be less than 1TB of SSD – is that still true or am I behind the times?
Yes, you are correct that 22.5K RPM drives were at one point one a futures roadmap back in 2000 and 2001 era, but never built due to all of the reasons layed out in these responses. The Power draw was excessive, the vibration was too high, the heat generated was over limit, the substrates were too brittle.
Once again, many great replies. Thanks all!
I looked up the StorageTek version of the 2305 and found a reference to the 4305 from the late 1970s. This sounds like a counterpart to the Dataram SSD in one of my earlier posts.
In reply to the question at the end of Peter’s post above, SSDs are about 10-20 times the price of a capacity HDD and are likely to remain that way.
I don’t know if this should be a separate thread but my comment about housing and spinning HDD vs SSD came out of a seminar on high density datacentre design by Dr Bob Sullivan (developer of the hot aisle/cold aisle model) I attended in London. He stated at one point that to get best rack density we should use SSD as it uses less energy and produces less heat. When challenged he immediately backed off and changed the subject. I always assumed that a TB of HDD capacity would be more compact and generate less heat than a TB of SSD but as I said I haven’t done a full analysis and wondered if anyone else has.
Yes Peter, SSDs generate less heat and consume less space per I/O than do HDDs, but they consume more space per GB than HDDs. They don’t absolutely HAVE to consume more space per GB, but they are more economical if they do.
Jim, I have to respectfully disagree with your assessment of SSD v. HDD pricing trends. My firm sells a good deal of SSD (both array and server side cache cards) and I can assure you that the price is much more competitive than you think. First, comparing “capacity” SATA HDDs to SSD in price isn’t exactly a level playing field. However, if you compare enterprise SAS HDDs such as Seagate’s Savvio 15K.3 (which is priced @ around $1.40/GB) to a solution from Pure Storage or Skyera (which are priced in the $3-5/GB range – complete) you can see the cost is closer to 2X or 3X over HDD ….for performance than runs circles around HDD. And even that comparison isn’t really fair to those SSD solutions as the SAS HDD price was simply a replacement drive w/o enclosure – whereas the SSD solution price is based on a complete solution that includes pricing for the enterprise array enclosure and all the supporting software (replication, snaps, thin provisioning, etc.) included. We also see prices coming down for SSD especially in the consumer SATA SSD cards where prices are in the 75 cents/GB range now (Look at the OCZ and Sandisk products).
I also have to question your comment on SSDs having to consume more space?…..why? OCZ offers the Z-Drive R4 that packs 16TB on one full height PCIe card. From an array perspective, TMS offers a 25 TB solution that is a 1U device for a standard 19″ rack…..not sure why you would think SSD should consume more space. Good discussion all.
Actually a lot of the latest innovation and gadgets that you see today on the marketplace are actually “old” technology
why I say this.
Back in 1999 I was a unix administrator at seagate singapore, back then if I remember correctly the latest disks on the market were running at around 5500rpm higher end disk were around 7500rpm but seagate already had 10k rpm and 15k rpm disk in their inventory not yet released to mass market.
Fast fwd to today 10k rpm is the norm and 15k rpm is the latest, so am very very sure they have 20k or 30k rpm in their inventory not yet released to the general market.
I can also tell you all the hardisk manufacturers and competitors are having similar inventory.
Why this is so?
You notice about gadgets whenever the latest gizmo is released to mass market shortly after other competing products also follow and make their debut, ever ask yourself how is it possible they all can do it almost around the same time?
My assumption is product development and marketing “cost” needs to be recovered so they keep selling the old stuff until it reaches certain market penetration or somebody releases “new product” then all also release their competiting products.
Well, in 2018, after visiting the Western Digital complex in California and seeing some of the new stuff they were working on, the spinning disk has hit its hey day back in 2012. That’s when they had that big flood, and many hard drives skyrocketed in price. It was also a great time to buy as well. Hell, the hard drive prices have not really been better actually. And I still have my 2TB and 3 TB drives from that time.
Funny how even today, the 2 TB is just about the standard today. the same cost comparison for larger drives took almost.. 8 years to catch up.. WOW. We are almost at the point where solid state storage is almost cheaper to make then spinning disks. Still, the spinning disk does have the cost per gb in late 2018. But as far as capacity? We still don’t have anything better then tape to store mass amounts of data in cheaply and long time.
Mickey, you’re right!
I have a chart somewhere that I need to find and dust off. It shows price per GB for HDD, Flash, and Tape, as well as floppies, CD-ROM, and Blu-Ray, that provide a good idea why the market has shifted the way that it has, and how it’s likely to shift in the future.
Thanks for the comment!
Jim
3D TLC NAND is giving platters a run for their money in space. Durability, probably not, though with wear leveling and people that aren’t throughput hogs, maybe they’ll last as long as a good platter?
At some point the strength of the material becomes an issue so there’s necessarily a limit somewhere. i don’t like the idea of a spinning HDD shattering and blasting through the computer’s case and into me.
That will happen to any metal given enough rotational speed. Spinning HDDs days are nearly over with the increasing dominance of SSDs which are just more reliable and have no moving parts.
This isn’t metal but polycarbonate, but it shows you what happens at 23,000 RPM to a CD. It warps due to resonant turbulence and then explodes.
https://www.facebook.com/theslowmoguys/videos/689408128510403/
A piece of thin aluminum at 30,000 RPM might not have much of a chance at survival. As the speed increases, the outward force increases beyond a linear factor, probably logarithmic or exponential.
Steve,
Thanks for the comment.
This is a really old post! You can’t even buy a new 15K RPM HDD any longer, they are obsolete, thanks to SSDs.
But I checked with Tom Coughlin, HDD analyst extraordinaire, (www.TomCoughlin.com) and he pointed out a couple of other issues:
– HDD companies tried to develop 20K RPM HDDs and found that “Flutter”, the bending and bowing of the disk, became too great at 20K RPM for a 2.5″ HDD, so they experimented instead with a 2″ form factor
– The storage capacity of a 2″ drive is really low, forcing the price per GB through the roof.
– The power of an HDD increases in proportion with the square of the spindle speed, so faster speeds consume enormous amounts of power.
– These three phenomena created market obstacles that prevented HDDs from going above 15K RPM 20 years ago when SSDs were too expensive for anyone but the military. Now that SSDs are affordable these market obstacles have simply become insurmountable.
Thanks for sharing the video. all I can say is “WOW!”
Jim