Hot Idea for a Faster Hard Drive

Ever see one of these? I bought on Ebay for no real reason and it sort of still works. I’m just about out of tape for it. A cleaner view is included as a stock footage clip that I hope someone will buy someday. I have had clips that have sat for a couple of years before anyone used them and this may be no exception.

http://www.youtube.com/watch?v=30x3vJz7Js8

http://footage.shutterstock.com/clip-1452379-stock-footage-antique-military-analog-test-signal-generator-running.html

I’ve often wondered why they never put two or four heads on a single hard drive platter.

We did it 30 years ago. There were four of them, and it boosted the speed 4X over a single one. But, even mainframe disk devices were really slow back then.

However, it's expensive. And you can accomplish the same thing with RAID-0 and multiple drives. A properly implemented RAID driver will read all the drives in parallel and assemble the results into a single data stream. Of course, each disk device has to be on independent SATA/SCSI/IDE channels.

The truth is that for most people, the seek and rotational latency is what really limits throughput. Seek latency is waiting for the head to move into position, and averages around 9 milliseconds. Rotational latency is waiting for the disk to spin into position, and averages 4.2 milliseconds for a 7200 RPM drive.

Compare that to how long it takes to actually move data. Drives with the latest SATA 6.0 Gbit/sec interface can sustain about 150 MByte/sec, but that is limited by the density of the bits on the drive platter.

The cluster size on your Windows NTFS file system isn't any larger than 32 kilobytes, and is likely smaller than that. Both the operating system and the disk drive do some read-ahead caching and will read more than you request at one time, but let's use 32K as an example. At a rate of 150 MByte/sec, a 32 KByte transfer would require only about 200 microseconds, or about 20 times the average rotational latency. You would have to read about 630 KBytes at one time to just split the average rotational latency time and transfer time 50/50, and that's not even considering the seek latency.

Are you seeing the problem? Very few people read a huge amount of data at one time, repeatedly, and would benefit from a faster sustained transfer rate. And those that do can construct a RAID array much cheaper.

Small random reads are much more common, especially on a computer with a virtual memory system. And that's why operating systems cache disk data in unused RAM, and even individual disks read-ahead and store data in their own internal cache RAM (which are about 32 MBytes these days). When you request that data, it gives you the cached copy instead, avoiding the seek and rotational latency delays.

Well, then its the seek time that is the problem and the value of the achievement this article announces is not what they claim. So do reduce seek time maybe the strategy should be to increase number of platters, decrease platter diameter, drastically increase the power of the hard drive motor, and dramatically increase the hard drive’s on board cache...or go the route of the hybrid SSD.

Thanks for the information/lesson. I honestly enjoyed reading it.

Thanks for the information/lesson.

You're welcome. I developed a disk driver many years ago, and always marvel at how far we have come in 3 decades. The magnitude of the numbers have changed, but the mechanical limitations are still the same.

To give you an idea of how much has changed, consider a typical consumer disk drive today: either a 3-1/2 inch desktop drive or a 2-1/2 inch laptop drive. The drive has its own cache and the drive controller circuitry built in. You can buy a moderate size drive for under $100, and hold it in one hand.

Back when I was doing disk driver development, the disk alone was about the size of a dishwasher. The disk controller (for up to 8 disks) was the size of a small kitchen refrigerator. And the cache memory was the size of a minivan (disk cache memory was rare in those days, though).

It isn't just disk storage technology that has advanced so far. Back then, a "scientific" mainframe was distinguished from a "data processing" mainframe by its ability to do floating point operations more quickly. If you were to run the same benchmarks on today's computers, you'd find that your iPhone is 20-50 times faster than the top-of-the-line scientific mainframes back then.

Well, then its the seek time that is the problem and the value of the achievement this article announces is not what they claim.

From what I read, it would be an improvement in sustained transfer rate, but I think the real benefit would be the increase in bit density. You would be able to cram more bits on the same sized platter, without increasing rotational speed. As usual, a journalist doesn't quite grasp the significance. :-)

So do reduce seek time maybe the strategy should be to increase number of platters, decrease platter diameter, drastically increase the power of the hard drive motor,

Decreasing the platter diameter would decrease seek time, but you would run into the bit density problem. Ditto with increasing the RPM -- it's already as high as 15,000 RPM for server drives. As the article points out, it takes time to flip a bit magnetically, and speeding up the platter just spreads out the bits. However, all of these things are expensive solutions, compared to alternatives.

and dramatically increase the hard drive’s on board cache...or go the route of the hybrid SSD.

I think that's where we should be going, and I'm surprised that we haven't seen more hybrid SSD's. There are a few, but they are limited in cache size, or are a completely new module, like this one:

http://www.engadget.com/2011/09/01/ocz-revodrive-hybrid-merges-100gb-ssd-with-1tb-hdd-for-499/

However, there are also laptop drives like this one:

http://www.newegg.com/Product/Product.aspx?Item=N82E16822148591

They appear to be a significant improvement in performance, but haven't really taken off. If I were replacing a drive in my laptop, I'd use one -- but I'm constrained to the standard offerings in my company's catalog.

SSD has to be implemented carefully: a single channel SSD is pretty slow to write. The good devices organize multiple SSDs into dual, quad, or even octo-channel configurations, in order to be able to sustain the kind of data rate you can currently get with a rotating disk device.

If you are using Windows 7 (and to some extent, Windows Vista), you can implement an SSD cache on your system. Get a fast USB thumbdrive, preferably USB 3.0 (both the device and your system). After formatting, right-click on the drive letter and enable "Ready Boost".

Windows 7 will use the SSD device for short reads. Longer ones will still go to the disk, because it can sustain higher transfer rates. Any writes are written to both the disk and the SSD device, so you can remove the SSD at anytime and your system won't crash. The data on the SSD is also encrypted, so there are no security issues.

If you have an SD card reader built into your laptop, that's an excellent way to use it. Just buy an 8-16G Class 10 memory card, put it in your laptop, and enable ReadyBoost. You can leave it there all the time.

If anyone decides to try this, follow-up and I'll post an explanation of how to use Windows Performance Monitor to see how much data is being cached, and how many of your disk reads are being satisfied by the SSD cache.

I've spent over 35 years in industrial electronics. The first "computer" I worked on was Westinghouse TTL logic. It was to control the continuous caster in South Works, Chicago, IL. The speed control was in a room off to the side of the control panel. NAND, AND, OR, XOR gates etc. were comprised of one board each. Each one was about 3 x 5 inches and plugged into a buss.

The whole room to control 4 lines of steel was about 10' by 15' and was air conditioned. Today I work on stacking machines run by PLC's that could run that entire system.

It's been an amazing increase in technology over the past thirty years. Even PLC's now are miniaturized.

Very interesting.

I think the point of a smaller diameter platter and a more powerful motor is to spin the platter from zero up to max RPM instantly. Smaller diameters have less rotational inertia. I believe this affects seek time.

Did you ever do anything with an Altair? I’ve heard their last version was sometimes equipped with a hard drive or two as add-ons in addition to a pair of 8” floppy drives.

I think the point of a smaller diameter platter and a more powerful motor is to spin the platter from zero up to max RPM instantly.

Rotating mass storage devices don't spin down, unless they are powered off. Some OS'es do so, if the power settings are set appropriately. Spinning up to RPM "instantly" is a massive amount of torque, even for the smallest disk, and it's not something you want to do repeatedly.

Rotational latency is determined by the speed of the disk, once it has stabilized. 4.2 milliseconds is the time it takes for a disk to rotate 180 degrees at 7200 RPM. That's where the "average" comes from: the length of time that you have to wait for a disk to rotate into position is -- on average -- 1/2 the time it takes to make a full revolution.

Did you ever do anything with an Altair? I’ve heard their last version was sometimes equipped with a hard drive or two as add-ons in addition to a pair of 8” floppy drives.

No, the stuff I worked on would fill a room, and heat several houses.