[guest]

I wonder why IDE, and SCSI, drives do not have a RAID 0 type arrangement internally. In this setup all read/write heads would be accessed in parallel. Think about it, you can't get any more syncronized than all the platters running off the same spindle. A two platter four head drive would have 2 times the transfer rate of a one platter two head drive. Since drives tend to fail mechanically entirely and not due to head/media issues this should not affect reliability.

When I was in college over a decade ago our class came up with this possibility for increasing data transfer rates. I normally believe "If you can think of it somebody is doing it" but in this case I guess not. I wonder why?

[Tea]

This question comes up every couple of months. And I post the same link every few months to answer it: http://www.redhill.net.au/d-o.html#2hp Essentially, the engineering challenges are huge, the potential performance gain is small. With the incredibly high track densities of today, it's probably not possible anymore anyway - not without making massive sacrifices in areal density and/or seek performance. Not worth doing.

[Virtual Larry]

Quote

This question comes up every couple of months. And I post the same link every few months to answer it: http://www.redhill.net.au/d-o.html#2hp Essentially, the engineering challenges are huge, the potential performance gain is small. With the incredibly high track densities of today, it's probably not possible anymore anyway - not without making massive sacrifices in areal density and/or seek performance. Not worth doing.

Hmm. Are you saying, that for multi-head/platter drives, once the seek is done and the heads have settled, that they *dont* read from all the data platters/sectors at once? I always thought that they did, although computing ECC, etc., for multiple parallel data streams at once might be computationally difficult.

[Mickey]

Correct. Current hard drives do not read from all heads at once, only one at a time. Same with writing. It's technologically possible to internal RAID, but not economically practical.

[freeborn]

Quote

Quote

This question comes up every couple of months. And I post the same link every few months to answer it: http://www.redhill.net.au/d-o.html#2hp Essentially, the engineering challenges are huge, the potential performance gain is small. With the incredibly high track densities of today, it's probably not possible anymore anyway - not without making massive sacrifices in areal density and/or seek performance. Not worth doing.

Hmm. Are you saying, that for multi-head/platter drives, once the seek is done and the heads have settled, that they *dont* read from all the data platters/sectors at once? I always thought that they did, although computing ECC, etc., for multiple parallel data streams at once might be computationally difficult.

I see Mickey has already answered. He is completely correct: current drives can only use 1 head at a time. After a head switch the actuator must perform a microjog to align the new head. Simple expansion/contraction over the operating temperature range of current drives cause the heads to move a few tracks relative to each other. It is not possible to have all heads on track at the same time over all temperatures the drive can encounter. In order to do it dual stage actuators or other more exotic (and expensive) technologies are necessary.

Free

[Virtual Larry]

Quote

I see Mickey has already answered. He is completely correct: current drives can only use 1 head at a time.  After a head switch the actuator must perform a microjog to align the new head.  Simple expansion/contraction over the operating temperature range of current drives cause the heads to move a few tracks relative to each other.  It is not possible to have all heads on track at the same time over all temperatures the drive can encounter.  In order to do it dual stage actuators or other more exotic (and expensive) technologies are necessary.

Free

Interesting stuff. I wondered about the track tolerances of modern drives.

Do you work in the drive industry by any chance?

[Mickey]

Speaking of dual-stage, that seems something the industry keeps talking about but never doing. It keeps getting pushed further and further out. While I was doing my senior project in college several years ago one group did theirs on piezoelectric technology with I think Read-Rite (or some other head manufacturer).

Even if you used that for internal RAID, you'd need one for each head. Goodness, that'd be a mess if IBM still had a five-platter platform! Servo control for 10 heads instead of 1 actuator.

[freeborn]

Quote

Quote

I see Mickey has already answered. He is completely correct: current drives can only use 1 head at a time.  After a head switch the actuator must perform a microjog to align the new head.  Simple expansion/contraction over the operating temperature range of current drives cause the heads to move a few tracks relative to each other.  It is not possible to have all heads on track at the same time over all temperatures the drive can encounter.  In order to do it dual stage actuators or other more exotic (and expensive) technologies are necessary.

Free

Interesting stuff. I wondered about the track tolerances of modern drives.

Do you work in the drive industry by any chance?

Thank you for the comment, I've been vacationing the past week and am just now catching up on this board. I try to keep informed on the IDE front and help when/where I can. My work? Best (for me) if I don't say, but thanks for asking.

Free