Hypothetical High-Speed Hard Disk Drive

[blazorthon]

I had an idea for a new HDD with vastly improved speeds hopefully rivaling SATA/SAS 6Gb/s SSDs.

A normal hard drive is has inside a set of (about) 3.5"/2.5" disks for storage. You can increase the data rate by increasing RPM or density but the fastest hard drives can only transfer data at 200MB/s. My idea is to replace a 3.5" set of disks with 4 disks just shy of 1.8" (1.8" is just a few millimeters to wide for 4 sets to fit in a 3.5" casing) and RAID 0 them internally like done in SSDs.

Now each individual platter may only be capable of about 100-155MB/s (To my knowledge only 3.5" drives reach speeds much over 155MB/s) but put together they may saturate a SATA or SAS 6Gb/s connection. The capacity could still go very high and probably be similar to a regular 3.5" drive.

This drive should have a higher cost /GB than a normal HDD but lower /GB than an SDD while being the same speed as other SATA and SAS SSDs. It also wouldn't suffer the slower and damaging writes of an SSD. If given current technology this drive may hold up to about 3TB with 4 sets of 5 platters at a maximum about (my estimation) 150GB per platter.

The only serious problems I can think of are high power usage compared to a single standard drive,
probably having a higher failure rate than a normal HDD because of its MANY moving parts (4 sets of platters and heads should give it a failure rate of at least 4 hard drives),


The power usage of a 2.5" drive usually peaks around 7 watts with 3.5" drives being roughly double. I would expect this drive to use about double the 3.5" but in high data density computers such as data centers (one place I would expect such a drive to replace standard 3.5" drives easily) it would probably have somewhat higher capacity and much higher speeds than the standard 3.5" 10k and 15k drives (I'm unsure of if this drive could hit 15k easily but 10k seems very likely). The capacity, speed, and power usage should just about equal 4 2.5" drives while being in a 3.5" size

The chances of any 1 of the sets of platters to fail may be higher than a standard HDD but the chances each 1 failing should be equal to a single drive failing. Also with 4 sets of platters there can be some very dynamic controls, hopefully allowing the other 3 sets to continue function should one fail.

My thanks to those whom took the time to finish my gargantuan post, I spent a lot of time contemplating and writing it.
Are there any problems I have missed?
Does such a drive seem plausible?

 

[Pyree]

If platter density and rpm are the same, then the disk performance depends on platter size only.

Capacity: diameter affects surface area- 1.8 inch x4 (10.18inch square), 3.5 inch (9.62inch square) not much difference in capacity.

Sequential speed: distance traveled per revolution depends on circumference of disk- 1.8 inch x4 (22.62 inch per revolution), 3.5 inch (11.00 inch per revolution) twice the speed, so looks good doesn't it?

Random speed: same as sustained speed? no, you can only seek the beginning of data from 1 platter. So, i.8 inch (5.65 inch per revolution), 3.5 inch (11.00 inch per revolution).

People want ssd because it has very low latency for good random read/write. Your new HDD design is going the opposite direction. Not to mention your HDD design will be hotter, noisier, more power hungry, more likely to crash, will be physically larger than 3.5 (already very cramp inside if you open one 3.5 and take a look inside, so I don't think you can fit 4x1.8 in there), cost more and takes more time to make, etc...

IMO, not worth it. 4x the platter number gives only the benefit of 2x sequential speed in a larger HDD format (I don't think the industry want to go back to larger 5.25 inch). So just raid a pair of 3.5, gives you the same performance as the 4x 1.8, and be happy.

 

[blazorthon]

The four platters would fit in a 3.5" case. Four 1.8" would be slightly to big to fit but 4 slightly smaller than 1.8" would fit. The point is to get a better speed per drive than a single 3.5". So it would give 2x sequential in the same form factor of 3.5". I already said the power would be a problem and I have an idea for the latency.

Thanks for your help I'll get back to work on it.

 

[RetiredChief]

Raid0 ONLY improves Sequencial read/writes. Raid0 does little to improve Boot time and program load times as these two operations are mostly dependent on access time (which in turn affects small file random read/writes) and Raid0 does NOTHING to improve this. What would be improved is editting large files such as video files (dvd/blueray a single file 1 to 40 gigs), large spreadsheets, cad/cam drawings and working with Large jpeg photos, and some gain in loading Map for games - NO improvement in performance once file is loaded.

The 4 x 1.8 In disks would have lower performance than just using 4 x 3.5 in disks.
Power for a desktop would not in itself be an issue, Approx 40 watts (<4 Amps) on a given rail - NO biggy. At best the 1.8 in HDDs might save a total of 10-> 15 watts. So Only advantage that I can see is total physical size, at the cost of reduced performance.

 

[blazorthon]

What if you had 2 read/write heads per platter? I realize increasing the moving parts isn't ideal but if you did it anyway would it improve random access speed to have a head on either side of each platter?

I apologize for not stating my goals in the first post. My goal is a drive with :
read and write speeds similar to or greater than Flash,
a non destructive read or write,
and lower cost /GB than flash.
Do you have a better solution?

 

[RetiredChief]

The only way to improve the small file acess time is to use raid0 with a short stroke.

Example. I used two 640 Gig WD black drives in raid0. Total available space is 1.28 TB. Then under creating the volume I selected 400 Gigs for size. You then leave the remainder of the drive vacant. In this case it decreased acess time fro approx 12.6 mSec to approx 9.5 mSec. (Google short stroke)

Note : cost was approx $100 for the 2 drives and 400 gig space so cost was much less than a SSD, But performance was still NO Wheres near an SSD.

 

[blazorthon]

My bad on the similar speeds to SSDs. It just needs to have similar sequential speeds to an SSD. latencies are intended to be similar to an HDD.

This is intended to be used as a high-speed drive such as all those expensive 300, 600, and 900GB drives. It would spin at least 10k RPM and possibly 15k RPM and it could cost as much as $400-$600. At those speeds it should have about 6mSec and 4mSec seek times with a single head per platter and hopefully double with dual heads.

It needs to have at least a similar random access speed to a standard 3.5" but I'm hoping for triple the sequential speed. In servers drives are often RAID 0, 1, or 10. Many current high-speed 3.5" drives use about 15-16 watts at peak and 2.5" are half that.

1.8" would probably be about half 2.5" in power usage but lets say my platters would use about 3/4 just to be safe. that gives it about 50% higher power usage than a standard 3.5" or roughly 30-35watts.

With two heads on (opposing sides of) each platter I would hope for similar latencies to that of a standard 3.5". Now I would expect triple the speed of a standard 3.5" being attainable.

About the reply on short strokes: If this drive can hold 3 TB than it has 5 times more data capacity than it needs (600GB) so I wouldn't mind cutting its space to 600-1200GB. That with two heads per platter could give it similar if not slightly better latencies to a standard 3.5" drive with triple the sequential data rate. Will this solution be plausible?

 

[fzabkar]

Hard drives already have 2 heads per platter -- top and bottom. ;-)

Do you mean 2 heads per SURFACE? If so, then this technology was used in the early 1990s.

ST12450W/WD Product Manual, Rev. B:
http://www.seagate.com/support/disc/manuals/scsi/8900b.pdf

Drive capacity, Mbytes unformatted 2,437
Read/write data heads, maximum (physical) 18
Read/write data heads, maximum (logical) 9
Internal data rate per physical head, Mbits/sec, 34.3 to 56.5
Internal data rate per logical head, Mbits/sec, 68.6 to 113.0

"The Barracuda 2, 2HP drive has two read/write channels on the circuit board. The SCSI-2 controller chip is modified to process data for two read/write channels at once. This technology effectively doubles the internal data transfer rate when compared to a conventional Barracuda 2 disc drive."

 

[blazorthon]

Yes that is what I meant. I hoped that with a set of heads on either side it would improve the latencies somewhat since the disc would only have to spin half as far for any single set of heads to reach a track. the sequential is not as important as the latency right now although if both improve I would be happy.

 

[blazorthon]

I've revamped the project. I will close this topic and prepare myself before opening under a new name so I'll thank everyone for their help on this here instead of in the new conversation. It could have taken me months to have gotten this far alone so thanks a bunch.

 

[blazorthon]

Best answer selected by blazorthon.