internally striped hard drives

[Shailesh Humbad]

With a RAID controller, one can stripe multiple hard drives for
performance. How about an individual drive that has internally
striped platters? It would have half the capacity, but (sometimes)
twice the performance. Does it exist?

 

[J. Clarke]

With a RAID controller, one can stripe multiple hard drives for
performance. How about an individual drive that has internally
striped platters? It would have half the capacity, but (sometimes)
twice the performance. Does it exist?

In a laboratory somewhere or as an antique piece of mainframe hardware
possibly. As a current-production commercial product, no.

If the heads couldn't move independently then the range of situations in
which there was any real performance boost would be very narrow.

 

[Folkert Rienstra]

With a RAID controller, one can stripe multiple hard drives for
performance. How about an individual drive that has internally
striped platters? It would have half the capacity, but (sometimes)
twice the performance. Does it exist?

So Shailesh, how many drives with fully independently movable heads
do you know?
Or do you even know what makes RAID0 tick?
Your question suggests that you don't.

 

[Ron Reaugh]

With a RAID controller, one can stripe multiple hard drives for
performance. How about an individual drive that has internally
striped platters?

A meaningless concept whatever you mean.

It would have half the capacity, but (sometimes)
twice the performance.

NEVER would give higher performance as only a single head is active at any
one instant.

 

[Bob Willard]

With a RAID controller, one can stripe multiple hard drives for
performance. How about an individual drive that has internally striped
platters? It would have half the capacity, but (sometimes) twice the
performance. Does it exist?

The IBIS drives of ~20 years ago used multiple R/W channels for
higher concurrent bandwidth, but that notion is not currently
cost-competitive: cheaper to add external RAID hardware (or even
software) to high-volume commodity HDs than to design HDs with
internal RAID and multiple R/W channels, since the low volume of
such RAIDed HDs would result in a very high price.

 

[Shailesh Humbad]

The IBIS drives of ~20 years ago used multiple R/W channels for
higher concurrent bandwidth, but that notion is not currently
cost-competitive: cheaper to add external RAID hardware (or even
software) to high-volume commodity HDs than to design HDs with
internal RAID and multiple R/W channels, since the low volume of
such RAIDed HDs would result in a very high price.

Seems like if there are multiple platters in a hard drive, then there
can be multiple, independent heads. Now with peer-to-peer PCI
Express, there would be plenty of bandwidth to support concurrent R/W.
In addition, hardware RAID chips get cheaper every day.

Even if the heads are not independent, there is lots of opportunity
for performance benefit, especially if the drive knows enough to put
contiguous blocks of data on parallel tracks. Of course, I'm assuming
that on current drives, even though the heads are touching all the
platters all the time, only one head is actively reading from a track
on a single platter at once. I don't know a lot about hard drives, so
I leave it to the experts here to correct me.

The reason I bring it up is that I would gladly take a 400GB drive
down to 100GB, if I can get a 4x improvement in performance for
sequential read/write. I only use 30-40GB of my drive now.

You can might say to just use four 100GB drives, but then I need a
bigger case, more power, and more cooling. My case is big enough to
hold that many drives, but I think having so many drives is
ostentatious, like a Lincoln Navigator. I'd rather have something
small and fast, like a Porche.

P.S. Most new types of computer hardware are expensive when they first
come out. That's not to say that I know there is a market for such
drives.

 

[Ron Reaugh]

Seems like if there are multiple platters in a hard drive, then there
can be multiple, independent heads.

Multiple heads, YES. One on each surface. There is only a single
servo/actuator assembly so all heads move in lock step. There is only a
single set of read/write electronics so only a single head may be in
operation at any given instant.

Now with peer-to-peer PCI
Express, there would be plenty of bandwidth to support concurrent R/W.

But no currently mfged HD that will do that.

In addition, hardware RAID chips get cheaper every day.

There's really no such thing as a HW RAID chip.

Even if the heads are not independent, there is lots of opportunity
for performance benefit, especially if the drive knows enough to put
contiguous blocks of data on parallel tracks.

Not feasible as only a single head can be in operation at any one instant.

Of course, I'm assuming
that on current drives, even though the heads are touching all the
platters all the time, only one head is actively reading from a track
on a single platter at once.
Right.

I don't know a lot about hard drives, so
I leave it to the experts here to correct me.

The reason I bring it up is that I would gladly take a 400GB drive
down to 100GB, if I can get a 4x improvement in performance for
sequential read/write. I only use 30-40GB of my drive now.

Making it smaller has nothing to do with RAID 0 like operation. RAID 0
doesn't throw away any drive space.

You can might say to just use four 100GB drives, but then I need a
bigger case, more power, and more cooling.

Nevertheless that's the way to do it. Four 100GB drives gives a 400GB RAID
0 array that potentially can read at four times the rate of a single drive.

 

[J. Clarke]

Seems like if there are multiple platters in a hard drive, then there
can be multiple, independent heads.

Such drives have been built in the past. The additional hardware including
motors to move each head ran the price out of the market. The last such,
IIRC, was a 2 gig Seagate made specifically for Cray.

Now with peer-to-peer PCI
Express, there would be plenty of bandwidth to support concurrent R/W.
In addition, hardware RAID chips get cheaper every day.

Regular PCI has plenty of bandwidth to support concurrent RW. The transfer
rates of individual drives are not all that high.

Even if the heads are not independent, there is lots of opportunity
for performance benefit, especially if the drive knows enough to put
contiguous blocks of data on parallel tracks.

That works fine until you need to add some data to the file and another file
has been started in the next block.

The only way I can think of that current drive technology could operate in
parallel would be if two or four or eight bits were written simultaneously.

Of course, I'm assuming
that on current drives, even though the heads are touching all the
platters all the time, only one head is actively reading from a track
on a single platter at once.

That is the way they work.

I don't know a lot about hard drives, so
I leave it to the experts here to correct me.

The reason I bring it up is that I would gladly take a 400GB drive
down to 100GB, if I can get a 4x improvement in performance for
sequential read/write. I only use 30-40GB of my drive now.

You can get some improvement just by formatting it to 100 gig. You'll be on
the highest-density section of the drive and also reduce your seek
distances and thus times considerably. Won't be 4x but it will be some
improvement.

You can might say to just use four 100GB drives, but then I need a
bigger case, more power, and more cooling. My case is big enough to
hold that many drives, but I think having so many drives is
ostentatious, like a Lincoln Navigator. I'd rather have something
small and fast, like a Porche.

I'd rather have something big and fast, like an SR-71.

P.S. Most new types of computer hardware are expensive when they first
come out. That's not to say that I know there is a market for such
drives.

I suspect that if there was a way to make multiple-head parallel reads and
writes practical someone would have done it. Given that each platter has
its own servo tracks these days I suspect that mechanical limitations make
parallel access economically if not technologically unviable at this time.

 

[Shailesh Humbad]

Making it smaller has nothing to do with RAID 0 like operation. RAID 0
doesn't throw away any drive space.

Oh yeah, you're right. Then what was I thinking? How about mirroring
the data on each of the hypothetical four 100GB platters? Then, we
make the four independently moveable, synchronous read/write, heads
write the same data to different cylinders in a distributed pattern,
e.g. cylinder 0latter 0, 16:1, 32:2, and 48:3. Then any particular
data should be on average four times closer to the nearest read head.
Writing might be slower though. I guess there are a lot of ways to
slice and dice the drive space, perhaps all of them impractical for
the mass market.

 

[J. Clarke]

Oh yeah, you're right. Then what was I thinking? How about mirroring
the data on each of the hypothetical four 100GB platters? Then, we
make the four independently moveable, synchronous read/write, heads

They we increase the price of the drive by some very large margin, if we can
even shoehorn in the mechanicals to move four heads.

write the same data to different cylinders in a distributed pattern,
e.g. cylinder 0latter 0, 16:1, 32:2, and 48:3. Then any particular
data should be on average four times closer to the nearest read head.
Writing might be slower though. I guess there are a lot of ways to
slice and dice the drive space, perhaps all of them impractical for
the mass market.

There wasn't enough demand for two-independent-head drives to keep them in
the market. What makes you think there's any demand for four?

 

[Folkert Rienstra]

Seems like if there are multiple platters in a hard drive, then there
can be multiple, independent heads.

That is even possible with a single (but dual sided) platter.

There are several ways to multiply the transfer rate in a drive without
RAID.

1) Interlace a cylinder: block 0 on side A, block 1 on side B block 2
on side A, block 3 on side B, etc. etc, reading 2 blocks simultaniously.
Obviously only works on dual sided platters and the problem is that
the tracks have to be so perfectly aligned that the servo only needs
to track one of the tracks or that the sliders will need to have a piezo
positioning element incorporated so that they can track individually.

2) Have 2 or more head-elements side by side on a slider, reading
as many tracks in parallel by a single head: block 0 on track 0, block
1 on track 1, block 2 on track 0, block 3 on track 1, etc. etc, again
reading 2 blocks (or more) simultaniously. This can even be done on a
single sided platter. Again there is the problem of alignment as only
one track at a time can be followed by the servo system. The slider
however has to sit on a linear tracking arm to avoid skew problems.

Both reintroduce the servo tracking platter approach which was
abandoned for good reason in favor of the embedded servo approach.

Now with peer-to-peer PCI Express, there would be plenty of
bandwidth to support concurrent R/W.

What has that got to do with a single 'striped' drive.

In addition, hardware RAID chips get cheaper every day.

What has that got to do with a single 'striped' drive.

Even if the heads are not independent, there is lots of opportunity
for performance benefit, especially if the drive knows enough to put
contiguous blocks of data on parallel tracks.

What has that got to do with a single 'striped' drive.

Of course, I'm assuming that on current drives, even though the heads

are touching all the platters all the time,

I certainly hope not.

only one head is actively reading from a track on a single platter at once.

I don't know a lot about hard drives,

You don't say. But you do about RAID?

so I leave it to the experts here to correct me.

The reason I bring it up is that I would gladly take a 400GB drive
down to 100GB, if I can get a 4x improvement in performance for
sequential read/write. I only use 30-40GB of my drive now.

What has that got to do with a single striped drive?
Since when do you have to sacrifice capacity to gain performance?
Are you on drugs or something?

 

[Alexander Grigoriev]

And one should not forget that the heads are never on the same track on
different platters, because of thermal misalignment. Keeping them on one
cylinder (whith a dedicated servo head) has become a big problem several
(drive) generations ago and caused many headaches because of need of
thermocalibration. Now, with track pitch 1/3-1/4 micron, it's all but
impossible.

 

[chrisv]

A meaningless concept whatever you mean.


NEVER would give higher performance as only a single head is active at any
one instant.

Clueless gibber.

 

[Ron Reaugh]

Such drives have been built in the past. The additional hardware including
motors to move each head ran the price out of the market. The last such,
IIRC, was a 2 gig Seagate made specifically for Cray.

ST12450W

 

[Ron Reaugh]

Oh yeah, you're right. Then what was I thinking? How about mirroring
the data on each of the hypothetical four 100GB platters?

RAID 1 throws away space for redundancy but RAID 1 doesn't increase the
drive's sustained transfer rate for reads.

 

[Folkert Rienstra]

Clueless gibber.

You think? Oops, stupid question.

 

[Folkert Rienstra]

RAID 1 throws away space for redundancy but RAID 1 doesn't
increase the drive's sustained transfer rate for reads.

Actually, it may for 2 files read simultaniously or for files that exist
of several fragments or that are bigger than 128kB as the individual
IO can be interspersed over the 2 drives.
Of course that doesn't work for writes.

 

[Shailesh Humbad]

ST12450W

"The Barracuda 2, 2HP drive has two read/write channels on the circuit
board. The SCSI-2 controller chip is modified to process data for two
read/write channels at once. This technology effectively doubles the
internal data transfer rate when compared to a conventional Barracuda
2 disc drive."

From: http://www.seagate.com/support/disc/manuals/scsi/8900b.pdf

Neat.

 

[Folkert Rienstra]

"The Barracuda 2, 2HP drive has two read/write channels on the circuit
board. The SCSI-2 controller chip is modified to process data for two
read/write channels at once. This technology effectively doubles the
internal data transfer rate when compared to a conventional Barracuda
2 disc drive."

From: http://www.seagate.com/support/disc/manuals/scsi/8900b.pdf

Neat.

But no mention of multiple, independent heads or motors to move each head.

Only dual R/W channels and 9 logical heads using 18 physical heads, 1 servo head.
Sounds like it is using a dual sided platter as a single logical surface, reading both
surfaces simultaniously.