Why dont large HDDs stripe the platters for parallel r/w ?

[Soubriquet]

As far as I know the large discs have the same IO speeds as the entry level. Yet none of them provide data fast enough to fill up ATA100 bandwidth, let alone SATA2.

I was hoping someone could explain to me why it is that the large multiplatter HDD do not read and write faster than single platter HDD.

In a large HDD there are 2 or 3 platters which means 2 or 3 read write heads. If they wrote data in parallel then data could be striped between the platters, kind of like a mini RAID array.

Why is this not done when it could make the large HDD 3x faster and use the wasted bandwidth ?

 

[velis]

Because electronics to enable that would be way more complex it is now.
Currently there's one data path for the head that's enabled. If you wanted to do that, you'd need more data paths + combining / splitting logic + buffering + whatnot. I believe that would faily complicate the actuator design.

 

[Bassyhead]

Separate actuators would be the least of the problems in a multi-actuator drive. The controller would have to be more robust, and a larger buffer would be needed. I would think 2 single actuator drives would still be cheaper than 1 drive with 2 actuators. Such a drive with, say, 2 separate actuators would need a controller that would be able to ready access to the next data on the next platter while the current head reads the current platter. It would be a very complicated design.

 

[Mark R]

The problem is that because the seeking is mechanical, and the cylinders on the platters so closely spaced - it would be extremely difficult to ensure that all the heads were over the same cylinder at the same time. Indeed, as temperatures changed and things aged, you alignment way well vary from day to day.

If it were done, it would likely make the head assembly much bigger and heavier - so seek times would be significantly slower than they currently are.

state-of-the-art drives it would be even more complicated.

In order to get maximum capacity, the platters are individually tested at the factory, and a low-level format designed specifically for each platter. So different platters, in the same drive, can have different bit densities and different cylinder spacings - depending on what particular format best suits each platter. This means STR, amount of data per cylinder, etc. can vary from platter to platter.

 

[EightySix Four]

That's what a RAID array is for is it not? You just want the raid to be inside the HD.

 

[Soubriquet]

sure, I cant see why not

I guess solid state is the thing to ami for anyway, but since everyone is trying so hard why not do this simple thing

if the reading head arms were a rigid array the only issue would be sequencing the data from the separate platters, thats just a bit of chipwork

 

[NeoPTLD]

It's already being done, the most common form called the RAID 0, which behaves like a series connected light bulb. Failure of one device would cause the entire system to fail and one out of two devices failing is statistically twice as likely as the only device failing in a single device system.

 

[velis]

Originally posted by: Soubriquet
If the reading head arms were a rigid array the only issue would be sequencing the data from the separate platters, thats just a bit of chipwork

No, sorry, it's not just the chipwork. If it were, it would already have been done. Like I said, you need more wires to accomplish that since each R/W head would need it's separate wires. I don't think that's how it's done now. +Some major chipwork reworking...

 

[Bassyhead]

Originally posted by: velis

Originally posted by: Soubriquet
If the reading head arms were a rigid array the only issue would be sequencing the data from the separate platters, thats just a bit of chipwork

No, sorry, it's not just the chipwork. If it were, it would already have been done. Like I said, you need more wires to accomplish that since each R/W head would need it's separate wires. I don't think that's how it's done now. +Some major chipwork reworking...

Yeah, you'd need almost a whole controller for each head, which all have to be integrated together. Can't use one controller for all the heads because that would defeat the purpose of having multiple actuators. It'd still be cheaper to buy separate drives.

 

[Soubriquet]

why would that defeat the purpose ? splain pls


if you had three arms running rigidly in parallel each could read one bit simultaneously, ditto write, so you write the data in parallel and read it in parallel, from the controllers perspective it is one platter with triple density.

 

[velis]

OK, Let's put it this way:
Drive performance significantly depends on seek time, not transfer rate. So if this was actually implemented, drive performance would go up a few percent, not 4x, 6x or whatever number of heads you have. There are very few uses that prefer transfer speed over seek time. Actually I can only think of audio / video editing right now.
Now, let's assume this is rather expensive since no drive manufacturer does it today. We have established, although quite unscientifically, that the market for such drives is rather small and that market uses RAID 0 anyway since they need capacity as well as speed, diminishing potential customer base even further.
Now, would you as a regular user be willing to pay $50 (pulled this figure from the top of my head) more for a drive that offers you a few percent performance increase over an otherwise identical, but cheaper drive? You can already do that by buying a Raptor. Do you have one?

As for defeating the purpose: I believe Bassyhead meant that having one controller would make all the actuators move exactly the same, not independently. This would not exactly defeat the purpose since you would still decrease rotational latency, but you would loose the benefit of being able to move the head that would obtain the results the fastest thus reducing the seek latency even more.

 

[Matthias99]

Originally posted by: Soubriquet
if you had three arms running rigidly in parallel each could read one bit simultaneously, ditto write, so you write the data in parallel and read it in parallel, from the controllers perspective it is one platter with triple density.

Alignment/formatting differences between platters (and even tracks within a platter) would make this nearly impossible to do reliably. Also, track density changes considerably from the inside to the outside of the disk, so if the heads were rigidly aligned, some of them would probably be unusable in certain zones.

Not to mention the significant increase in the complexity and weight of the read/write head assembly. This would require bigger, hotter, more power-hungry, and more expensive actuators. Besides driving up prices, that would probably hurt reliability, and space is pretty tight in a 3.5" HDD already.

I've seen this suggested several times before just in this forum. This is one of those ideas that sounds simple and great in theory, but is very difficult to implement physically. At current prices, you're far better off just using a striped RAID array if you need an increase in STR.

 

[tk11]

I think what velis said makes alot of sence. In which case wouldn't having 2 heads/platter be like doubling the rpm of the drive if they were positioned opposite each other? Or if one head was always positioned over the inner tracks and the other over the outer reducing the distance the heads would need to move.

 

[velis]

Originally posted by: Matthias99
Alignment/formatting differences between platters (and even tracks within a platter) would make this nearly impossible to do reliably. Also, track density changes considerably from the inside to the outside of the disk, so if the heads were rigidly aligned, some of them would probably be unusable in certain zones.

What I said makes sense, but read Mathias99's post. It's practically impossible to do. Sometimes the drive "looses" it's alignment data and starts "generating" bad sectors. A low level format will fix such problems since alignment data is written anew to the platters.
And this is just with one actuator. I think aligning two actuators would be nearly impossible. It would however be possible to have one for outer and another for inner tracks and that just might shave some 3-4 ms off average synthetic seek time. But since the seeks are usually a much smaller distance than half of the disk it wouldn't really help in real-life situations.

Keep in mind that having multiple actuators is virtually impossible to do. On the other hand it is possible to have the heads on one actuator do their stuff in paralell effectively creating a RAID0. At least I cannot see any reason for it to not be possible. It's just too expensive to do.

 

[Mark R]

Originally posted by: velis

What I said makes sense, but read Mathias99's post. It's practically impossible to do. Sometimes the drive "looses" it's alignment data and starts "generating" bad sectors. A low level format will fix such problems since alignment data is written anew to the platters.
And this is just with one actuator.

Alignment data is written to the drive at the factory with an ultra-high precision servowriter machine. The drive mechanism is not precise enough to rewrite the alignment data itself - hence a 'low-level format' is just a fancy name for a data wipe. A true low-level format has been impossible since IDE was invented.

I think aligning two actuators would be nearly impossible. It would however be possible to have one for outer and another for inner tracks and that just might shave some 3-4 ms off average synthetic seek time. But since the seeks are usually a much smaller distance than half of the disk it wouldn't really help in real-life situations.

The actuators are, too an extent, self aligning - because all actuator movements are done with reference to the alignment data. During an actuator movement, the heads pick up the signals as they cross over the alignment cylinders - the drive controller then accelerates/brakes the actuator (watching its position as it goes and making minor tweaks to the trajectory) so that it lands as close as possible to the desired cylinder. If it misses, well, it just tries again - although most drives will record this event and if they occur often, will probably flag a SMART warning.

On the other hand it is possible to have the heads on one actuator do their stuff in paralell effectively creating a RAID0. At least I cannot see any reason for it to not be possible. It's just too expensive to do.

Actually, this would be essentially impractical.

Modern drives pack the data so close that the cylinders can be less than 1/10000 inch apart. Any minute misalignment - in the platter spindle, or the the actuator mechanism will make parallel reading from multiple heads on the same actuator impossible. All bearings have some degree of play and will wear - so even if you did get it right at the factory it may not last very long. Even temperature changes as the drive warmed up could cause uneven expansion/contraction of the actuator assembly, causing hopeless misalignment.

Additionally, the cylinder spacing can be different from platter to platter - or even from one side of the platter to the other. The manufacturer will test each platter individually, and depending on how the magnetic coating turned out will select a different low-level format for each zone on each platter.

 

[velis]

Ooops, I always understood the factory written alignment data was sort of an approximate supplemented by additional fine-precision data written during low-level format.

I checked here: http://www.storagereview.com/guide2000/ref/hdd/geom/formatLow.html
and you are abolutely correct.
Of course all the other points deal with the fact that the base assumption was incorrect

I stand corrected.

 

[orangat]

I think people still think they can low level format their drives because of manufacturer sw utilities (like Maxtor's) which have a llf description although they don't actually write tracks and sectors and just do a zero fill.

 

[drinkmorejava]

Ummm, last time I checked, multiplater drives do stripe internally.

 

[Oxides]

A few years ago someone was making cd drives with multiple, independent heads to do just this.