Ok, please disregard if this has been covered or is just a plain stupid idea.

How come HD manuf. have not created a single drive w/raid-0. What I am trying to say is, you have 2-5 platters in a disk. What if you were to have 2 platters w/ 100 GB each in the drive, but instead of having 200GB total, they raid the two platters within the drive and thus giving you a single drive raid and at the same time being invisible to the OS. AND what if you had this single drive raid and then made an actual two disk raid and thus having Double raided 0 drives........ If that makes any sense at all. LOL.

Comments??? Or is this idea too far out or maybe already have been done.

Mike

The problem is that config would need to readers (one for each platter), and that would skyrocket the price of the disc, wich is bad

Also including the raid controller on-disc also would increase prices.

The disc would have less capacity as well because of some space being used for other things instead of platters.

it would be faster and simple but it wouldn't be as convenients as typical Raid-0.

The future I think lies in Hybrid drives and with higer densities (¿I think I misspelled?) per platter.

Maybe a SATA limitation? A RAID setup will have one SATA cable for each disk, while what you're thinking of would only have one cable. No idea if that matters - Let me see, 3 Gbps/s is about 375 MB/s and that's several times what a good disk needs.

Another guess: there are lots of RAID types and disk types and sizes and it's probably best to let the user decide what type he wants and buy the required number and type of disks and so on. Plus, if a disk dies you can replace it individually, you don't have to throw everything out.

Maybe heating: several separate disks in RAID may have less of a heating issue than the same disks enclosed in common metal box.

Just guessing of course. I like your idea. I'll buy that thing if they make it

Good point, but the read/write heads should not need to change as they would all move in unison as they already do. As far as the extra raid logic, things would come down in price if they would make it standard for all drives.

I understand that flash drives will become the future BUT the storage densities are not there yet and still cost an arm and a leg. Hybrid is promising but not fast enough.

Anyways, thanks for the comments, I am just throwing ideas around.

Mike

**Edit** sorry wanted to quote - supremelaw


WOW, thanks for the info. Even with the example that you provided with the 15k drive reading at 125mbsec, yes thats fast but you can make it faster by incorporating the single drive raid alongside that and theoretically doubling that speed.

Mike

I'm pretty sure this is already done because the armature is a single unit and the arms do not move independently between platters and I read somewhere that they did write the drivers on multiple platter drives to read/write on multiple platters simultaneously.

If you notice the armature is on one shaft, and there are no joints on the arms themselves. When the armature moves, all the arms move.

What I was after was just one armature and two heads reading at the same time on two different platters.

But if two servos are required to operate the raid, then it still wouldnt be half bad.


Mike

Interesting idea. I think though, that we're going to see bigger and bigger single drives for capacity, and faster and faster SSDrives for speed. We'll see. When SSD is down to maybe $1.50-$2.00 per GB, I'll probably buy at least one.

I believe the limitation for read/write to different heads simultaneously lies in the controller and circuitry architecture.

Currently, a single channel controller chip controls everything, and does so one head at a time. As soon as one track on one platter is read, the controller electronically switches to another head and begins reading from there. LBA addressing is handled in sequential fashion.

The channel controller chip does a huge amount of processing on reads/writes due to the encoding and error correction of writing/reading data on the platters. To do reads/writes from more than one head at a time, you would need multiple channel controller chips, which raises costs. Then, on top of that, you need a 3rd controller chip to manage and synchronize the two. The 3rd controller chip is essentially a RAID controller, and would do things like buffering to/from cache, and performing the SATA interface piece.

Thus, you're talking about a redesign of the entire hard drive's controller board. Instead of the current channel controller chip, which performs several major functions all in one chip: (error correction/encoding/decoding, read/write gain control, servo positioning, head movement, cache management, and SATA protocol), you now have to split the tasks up:

Channel controller chips: Error correction/encoding/decoding, read/write gain control. - Need one for each head or group of heads for user data, possibly another one dedicated to servo positioning reads (remember that servo positioning data is located on one side of one platter, must be read continuously to correctly position the heads).

Master controller chip: Servo positioning, head movement, cache management, RAID function/interleaving, SATA protocol

Cache: Probably need to increase cache to take advantage of grouped writes to minimize head movement.

It may even be difficult to get this amount of logic onto a single HD controller board. It's a pretty steep increase in circuitry.

The reward would be a hard drive with massive STR. With a 2-platter drive (4 heads, 3 for user data), you could conceivably get close to 300MB/sec reads and writes if other current technologies (perpendicular recording) were also used.

Please note: I don't work for a hard drive company, so what I know of hard drives is only superficial study and what I've read. I could be wrong about any number of things. But to the best of my ability, this is the way I understand the drives and the engineering involved.

I'm sure it could be done that way. Would it be worthwhile for the extra cost? Not too likely. Current OS's such as Linux (can't be sure of Windows) retain as much information in their memory cache as they can. If you have 4 gigs, probably around 3 gigs of the last reads would be cached. With the current cost of ram and that kind of caching, disk read speed becomes almost irrelevant for most applications. For example I can start a compile on a 2 core CPU with 2 gigs of RAM running 4 threads. The compile will peg both cores, and disk IO is very light. An old 30 gig keeps up just fine.

Database R/W performance is a different animal. By the time R/W performance is an issue, cost is usually not a determining factor so traditional RAID solutions are employed.

Well if you had two armature assemblies, you could theoretically hook two SATA ports up on the same HD. In that way, it would still be a RAID 0 setup and should be cheaper than two physically separate drives. Furthermore, you wouldn't be using any more armatures, you would just need a separate servo and control system for it. It would be an interesting solution and should be cheaper than two drives. If you had a space-conscious setup, it would be the way to go (until SSDs anyway).

The major flaw here is that it is still a mechanical solution. SSD is the wave of the future. Before too much longer we will have SSD that is as fast and cheap as RAM. IMO, Hybrid drives are a misguided stop gap solution, in an attempt to milk the last drop out of the mechanical cow. Most likely, the kings of SSD will not be Seagate, WD, Toshiba, or Hitachi. They will be chip/memory makers.

From what I read it will be sooner than later. Maybe within 2 years. It seems that the chip fabs only require "modifications" rather than all new machines. Current chip guys like Micron, Samsung, etc... will probably be the first as they seem to have the deepest pockets. Even still they will probably want a healthy ROI initially.