Dual Heads for Hard Drives?

[CplHicks]

I was thinking about this a few days ago and was wondering if it was possible. Could you have two heads on a hard drive, say one for reading and one for writing (or both for either)? It would seem like this is possible since you could position them on opposite sides of the magnetic disk and make sure that they didn't read/write the same thing at the same time. Has this ever been tried or is it not possible for other reasons?

 

[naukkis]

It can be done but there's no benewiths, only handicaps in both performance & reliability.

 

[Locutus4657]

Hard drives already have multiple heads, one for reading each platter. They are all linked together though.

Carlo

Originally posted by: CplHicks
I was thinking about this a few days ago and was wondering if it was possible. Could you have two heads on a hard drive, say one for reading and one for writing (or both for either)? It would seem like this is possible since you could position them on opposite sides of the magnetic disk and make sure that they didn't read/write the same thing at the same time. Has this ever been tried or is it not possible for other reasons?

 

[dkozloski]

What you are describing is a RAID 0 setup in one case sharing platters; an interesting concept until one head tries to go past the other.

 

[zephyrprime]

It's been done before. I remember Western Digital had one but I don't think they ever sold it. The heads were positioned on opposite sides just like you said.

My roommate came up with the idea of a head "wing". Instead of a single little slider, there would be a bar that spanned the radius of the platters. There would be numerous read/write heads and the head would be push/pull actuated instead of axially mounted. Who knows if it would work.

 

[Smilin]

There was some research done on it in the 80's involving up to like 4 heads per platter side I think. Never heard what came of it...no actual product obviously.

I think probably the easiest application would be to have two extra heads that simply read the tracks on either side of the one you are on into cache. A performance gain should been seen since (barring nasty fragmentation) you are most likely to read the next track in sequence if the current track doesn't meet your needs.

Another head dedicated to being near the FAT/Directory tables would be nice too. Heads spend alot of time flopping about writing data, updating entries, writing data etc...

 

[ProviaFan]

Two heads, one on either side of the hard drive, might increase performance if you're running two partitions on one drive. However, it would be difficult to implement, and probably quite expensive too, so IMHO it's not worth it.

 

[Souka]

and most importantly, more parts = higher complexity = higher cost.

follow the KISS rule.....

 

[CplHicks]

Originally posted by: jliechty
Two heads, one on either side of the hard drive, might increase performance if you're running two partitions on one drive. However, it would be difficult to implement, and probably quite expensive too, so IMHO it's not worth it.

I agree that it would increase costs, but by how much? Hard drive prices have dropped quickly if you look at GB/$ and drive capacities continue to rise. However performance is still relatively weak, extremely slow compared to the rest of a modern PC. Personally I'd sacrifice a few GBs of space for a faster performing drive at the same price.

 

[JustinLerner]

I'm unfamiliar with the way current hd's read/write. Do they read/write across multiple platters at the same time to increase overall I/O throughput rates?

While traditional HD head arrangements utilize pivoting arms, I noticed that CD-ROMS/DVD-ROMS have a different linear path for optical head movement. CD devices have access times around 70+ms and access time would need improvement for this type of head mount use in HD's.

Using head mounts like the ones used in CDs, couldn't one easily have independant heads with the necessary degrees of separation between head supporting arms [ON THE SAME PLATTER] and even apply this methodology to different platters? This could effectively [ALSO] allow the same drive to support multiple, simultaneous reads and writes from independent heads on different faces of each platter if the bus could handle the different I/O streams in parallel. The harddrive bus connections would have to be better than SCSI variants and more like multiple, parallel SSA, or one SSA channel for each head.

 

[Smilin]

HDD's already use multiple heads in the manner you speak of. Each platter has two heads one for the top and one for the bottom. The top and bottom tracks are read together and referred to as a cylinder. A two platter drive for instance has four heads that work together to read a single 'cylinder' that extends through all four surfaces.

The question of the thread is more along the line of having more than one head per surface.

 

[JustinLerner]

Smilin, you didn't answer my first question, and yes I understand that multiple heads on different platters move in unison, but do they all read/write in unison across multiple platters together? Also if you read the first post in the thread you will notice that multiple heads on the same platter are the initial topic.

I describe multiple heads on the same platter AND additionally also on different platters working independantly of one another using a the type of head mount and arm that is used by CD devices, not typical HD's. If you understood me properly, this would allow two heads or more per surface AND/OR platter and permit independant head operations.

The advantages of being able to read and write from different areas simultaneously may far outweigh the 'speed' of single reads or writes for applications and systems that have heavy database use or users running multiple programs.

----

This concept could even be applied to optical media like CD/DVD-RW media, allowing multiple read and write access simultaneously.

 

[ChefJoe]

There's a reason that 1 platter HDs cost a minimum of $40... doubling the head cost while decreasing teh production (specialty line) makes that even worse.

 

[Peter]

HDDs do have multiple heads, one per surface. However they use only one at a time. Parallel use of N heads would also mean having N times as much read/write analog circuitry. These are expensive. Really.

The very first Seagate Barracude high performance SCSI drive series actually had twin sets of the entire read/write hardware. One set of heads reaching into the platter stack from the front, and the other one from the opposite side. The drives were longer than usual ones, and also twice as expensive. They were incredibly fast (for the time) in random access time though.

regards, Peter

 

[Pariah]

A few years back Conner developed a drive that had two actuators and 2 sets of heads (what the original poster is asking about) called the Chinook. The company was subsequently bought out by Seagate and I believe they were the ones that actually released the drive. Problem was, it was so complex and the performance gains weren't that great, plus the added cost was unwanted, so the technology was cancelled.

"but do they all read/write in unison across multiple platters together?"

No, only one head can read/write at a time.

"CD devices have access times around 70+ms and access time would need improvement for this type of head mount use in HD's."

CD's have horrible access time due to the inconsistency of the media. Different discs use different reflective layers with lower or higher reflective properties, scratches, dust or fingerprints on the disc, also the discs wobble differently often greatly at high RPM's, which all add up to a much tougher surface to focus on than a hard disc.

"Using head mounts like the ones used in CDs, couldn't one easily have independant heads with the necessary degrees of separation between head supporting arms [ON THE SAME PLATTER] and even apply this methodology to different platters? This could effectively [ALSO] allow the same drive to support multiple, simultaneous reads and writes from independent heads on different faces of each platter if the bus could handle the different I/O streams in parallel. The harddrive bus connections would have to be better than SCSI variants and more like multiple, parallel SSA, or one SSA channel for each head."

Optical discs and HD's use completely different methods for reading the media. Optical drives use a laser to reflect of the discs detecting pits which alter the reflection. Hard discs are magnetic media read by heads that detect the magnitism of the cluster on the disc. The two technologies have nothing in common that can be applied to the other.

 

[Gunbuster]

<<HDDs do have multiple heads, one per surface. However they use only one at a time. Parallel use of N heads would also mean having N times as much read/write analog circuitry. These are expensive. Really.>>

The other problem is heat, platters and pickup arms expand as the drive warms up, you have no way to make sure 2 or more heads are tracking correctly on the same arm

 

[SCSIRAID]

A variation of dual actuator (which is what it seems you are describing - where each platter has two heads per surface but independantly positioned) is split actuator. The split actuator eliminates the thermal tracking problems which dual actuator brings. In a 4 platter drive - split actuator means that 2 of the platters are covered by a single actuator (1 head per surface) while the other 2 platters have ANOTHER single actuator with 1 head per surface. This approach greatly improves the drive performance in small block random workload (typical of commercial servers). It also would allow some RAID 0 effect for very large reads (if you interleave the head numbers). The downside is cost and complexity since you would need dual positioners/servos and dual read channels. I dont believe demand will be high enough to drive actual products.

 

[Den]

This has been done in mainframe class hard drives as well. (DASD for you IBM dudes). Even there though where cost is not really an issue and performance is king, this has been phased out. As some have said, it worked by haveing two full sets of heads and arms, each activiated by it's own actuator, one on each side of the drive. The performance was just not enough to justify the cost, the cost was much more than double (due to the much smaller quantity produced) and the performance was much less than double. It winds up being cheaper and faster to do some variety of RAID and use two drives.