Super-performing multi-actuator arm hard drive

[Vaqueropp]

If you like the HD’s to compete with current SSD’s, at least for the short tem, you may want to have a double, triple or quad actuator arm in a small 2.5 factor disk. With a smarter controller, you could have the benefit of deciding to use them all for reading, writing, or  a combination of reading and writing at the same time.

Rotational Latency will be divided between the arms

Translational latency will be minimized if the design allows for smart predictive landing zone of each arm - whichever is closer to the data can access it.

Putting various arms like this will probably behave better than current RAID arrangements, since the internal architecture will not be limited to a static RAID layout (it’s the same physical disk!), but will be smarter since it will know when all heads will be reading (possibly from same cylinder if necessary), writing or doing both and deciding when the arms should execute the operation.

This does not change the current technology at the low level: it merely parallelizes a proven solution, and makes it smarter.

If you add a bigger buffer to the disk, it will be hard to beat.

[MRFSYS]

  If you like the HD’s to compete with current SSD’s, at least for the short tem,

  you may want to have a double, triple or quad actuator arm in a small 2.5 factor disk

We asked a very similar question of WDC many years ago,

and we even found patent drawings of this very thing.

WDC’s Customer Support replied promptly, and professionally,

to say that a dual-servo HDD would be much more expensive

to manufacture, and it would also be much more prone to fail

simply because of the larger number of moving parts.

And, believe it or not, the solution is really staring us in the face:

use 2 x HDDs in RAID 0 arrays, if you want 2 x actuator arms!

Along the same lines as WDC’s reply, the industry was also

exploring much success with short-stroked partitions

particularly after perpendicular magnetic recording became

the standard for new rotating platters.

Here is a very revealing graph which compares the

effects of PMR and short-stroked partitions:

http://www.supremelaw.org/systems/io.tests/platter.transfer.crossover.graphs.2.jpg

Not only does PMR allow greater data density on any given track,

it also allows tracks to be much closer together: 

this explains the wide differences in the slopes of those graphs.

And, it should be immediately apparent that short-stroked partitions

will necessarily perform much better over time:

just compare track-to-track access times

with “full-stroke” access times in the factory’s specs.

This property of rotating HDDs is easily explained:

the recording density on each track must remain

relatively constant from outermost to innermost tracks,

in order to simplify the logic of the magnetic read heads.

Thus, there is necessarily much less data on the

innermost tracks, given a relatively constant recording

density.

And, the amount of raw data on any given track

is directly proportional to track circumference,

computed as Pi x Diameter, or 2 x Pi x Radius.

You can measure and graph this property yourself,

using the popular HDTune software / Benchmark tab.

I hope this helps.

MRFS

[Kramy]

The real answer to this problem lies in a RAID-1 implementation that treats the second drive as a second head. If one head can stay doing one task while the other does everything else, the amount of seek time wastage would drop dramatically, vastly improving read performance on one drive. During some tasks 80%+ of the time is spent seeking rather than reading, and when multitasking heavily it could be even higher. A second head would do wonders in situations like that.

If you combine this with lazy write buffers (writes must reach the primary drive, but can hit the other one whenever), then you’d have quite incredible performance scaling on your hands. RAID-0 is good for video work, but this kind of RAID would be incredible for multitasking, gaming, etc.

You wouldn’t be able to call it RAID-1, though. RAID-1 is for data security… and anything involving lazy writes is NOT for data security. Maybe call it RAID-1G/RAID-1GLW or something. RAID-1 to indicate the mirroring effect, G for Gamer (gamers wouldn’t care about security at every second; a gamer may worry about having a mirror in the event one drive dies, but an up-to-the-second mirror is not a priority; and if the computer ever BSODs, the primary drive can be used to reconstruct the lost writes on the secondary drive during bootup), and LW for the lazy-write aspect. (So RAID-1G would split the reads as efficiently as possible (for multitasking) to minimize seek times, but mirror writes in real-time. RAID-1GLW would have lazy-writing on one drive, and obviously perform better.)

Heh… I wouldn’t want to be the coder that has to implement that RAID level.