Single hard-disk mirror design idea

[zoiled]

Hi!

I just snooped around the net for information on how hard-drives work
(in detail). I found (if the information is up to date) that all the
read/write heads are moving as one single unit. Also I read that when
reading or writing the hard-drive usualy only use one head at the time.
Finaly I read that one side of each plattern is used for position
information.

So here's my idea. If the most reliable way to manufacture a read/write
heads is to make all heads move in unison why would you want to
read/write with only one head instead of distributing the data over all
heads. (just like a RAID array stripes data for example).

To overexplain this let's say you have a hard-drive with 8 platterns
and 8 read/write heads. If you write one byte (8bits) of data each bit
would be "assigned" it's own plattern making it possible to write one
byte with one "pulse". instead of "pulsing" this byte on one plattern.

I guess someone have allready had this idea but why isn't it smarter
for a hard-drive to internaly stripe the data over all platterns
instead of writing it one by one. Is it perhaps some reason who makes
it physically impossible for all read/write heads to do a read/write
all at once?

 

[Rod Speed]

(e-mail address removed) wrote

Hi!
Lo!!

I just snooped around the net for information on how hard-drives
work (in detail). I found (if the information is up to date) that all
the read/write heads are moving as one single unit.
Correct.

Also I read that when reading or writing the
hard-drive usualy only use one head at the time.
Correct.

Finaly I read that one side of each
plattern is used for position information.

That hasnt been true for a long time now.

So here's my idea. If the most reliable way to manufacture
a read/write heads is to make all heads move in unison

It isnt done for reliability reasons.

why would you want to read/write with only one
head instead of distributing the data over all heads.

Basically because there is only one set of electronics
that is doing the reading and writing thru a particular head.

(just like a RAID array stripes data for example).

Nothing like, actually. The short story is that it makes
a lot more sense to have two physical drives instead
of being able to read and write to two heads at once.

With RAID1 you dont lose everything if one drive dies.

With RAID0 you can independantly seek the two physical
drives and use the one that gets there first etc.

To overexplain this let's say you have a hard-drive with 8 platterns
Platters.

and 8 read/write heads. If you write one byte (8bits) of data each bit
would be "assigned" it's own plattern making it possible to write one
byte with one "pulse". instead of "pulsing" this byte on one plattern.

It wouldnt be done like that if you could use all 8 heads simultaneously.

I guess someone have allready had this idea

Its just not practical. If you do need better speed, it makes a
lot more sense to use ram instead, and that is what is done.

but why isn't it smarter for a hard-drive to internaly stripe
the data over all platterns instead of writing it one by one.

Basically because you would need 8 times the read write electronics

Is it perhaps some reason who makes it physically impossible
for all read/write heads to do a read/write all at once?

Nope, its possible to do it that way.

 

[Daniel Lang]

So here's my idea. If the most reliable way to manufacture a read/write
heads is to make all heads move in unison why would you want to
read/write with only one head instead of distributing the data over all
heads. (just like a RAID array stripes data for example).

To overexplain this let's say you have a hard-drive with 8 platterns
and 8 read/write heads. If you write one byte (8bits) of data each bit
would be "assigned" it's own plattern making it possible to write one
byte with one "pulse". instead of "pulsing" this byte on one plattern.

I guess someone have allready had this idea but why isn't it smarter
for a hard-drive to internaly stripe the data over all platterns
instead of writing it one by one. Is it perhaps some reason who makes
it physically impossible for all read/write heads to do a read/write
all at once?

In the old days when disk drives had a single dedicated servo track,
it was possible to do this. The cost of the additional electronics
made it cost prohibative for the most part.

Nowadays, the cost of the additional electronics is much less.
The problem now is that to handle the higher track densities,
disk drives have embedded servos on each surface of each platter.
This allows the manufacturers to overcome differential expansion
and contraction between the heads but only one head can be brought
into alignment at a time. This restricts reading and writing to
one head at a time.

Daniel Lang

 

[Waxhead]

(e-mail address removed) wrote


That hasnt been true for a long time now.


It isnt done for reliability reasons.


Basically because there is only one set of electronics
that is doing the reading and writing thru a particular head.


Nothing like, actually. The short story is that it makes
a lot more sense to have two physical drives instead
of being able to read and write to two heads at once.

With RAID1 you dont lose everything if one drive dies.

With RAID0 you can independantly seek the two physical
drives and use the one that gets there first etc.


It wouldnt be done like that if you could use all 8 heads simultaneously.


Its just not practical. If you do need better speed, it makes a
lot more sense to use ram instead, and that is what is done.

I can't image that *just* more ram will give incredible speed. You will
eventually end up waiting for the ram to be flushed to disk anyway.

Basically because you would need 8 times the read write electronics


Nope, its possible to do it that way.

Thanks for some excellent answers

 

[Waxhead]

In the old days when disk drives had a single dedicated servo track,
it was possible to do this. The cost of the additional electronics
made it cost prohibative for the most part.

Nowadays, the cost of the additional electronics is much less.
The problem now is that to handle the higher track densities,
disk drives have embedded servos on each surface of each platter.
This allows the manufacturers to overcome differential expansion
and contraction between the heads but only one head can be brought
into alignment at a time. This restricts reading and writing to
one head at a time.

Daniel Lang

So what you are saying is that each head moves independently from the
others but only a single one can be used for a read/write operation?
Isn't all heads moving as one single unit?

Anyway how about two (or more) "head units" mounted inside a
hard-drive. Each unit taking care of each section of the platter. One
unit from center to middle and another for middle to the end. Also if
each unit was designed to have the possibility to move over the entire
platter it could have functioned as a "failsafe" in case one of the
heads was failing.
....but then again I guess someone else allready had this idea

btw: thanks for your answer

 

[craigm]

Waxhead wrote:

So what you are saying is that each head moves independently from the
others but only a single one can be used for a read/write operation?
Isn't all heads moving as one single unit?

Anyway how about two (or more) "head units" mounted inside a
hard-drive. Each unit taking care of each section of the platter. One
unit from center to middle and another for middle to the end. Also if
each unit was designed to have the possibility to move over the entire
platter it could have functioned as a "failsafe" in case one of the
heads was failing.
...but then again I guess someone else allready had this idea

btw: thanks for your answer

You are missing the point about thermal expansion. When the head on one
surface is above the desired track, the heads on other surfaces may be
tracks away from the desired location. Even when switching heads, the drive
has to seek to the correct track.

Your second idea requires a second headstack assembly. This would add
significantly to the cost of the drive, including duplication of most of
the electronics.

Disk drives are commodity items. Anything that raises cost to manufacture is
an issue.

craigm

 

[J. Clarke]

So what you are saying is that each head moves independently from the
others but only a single one can be used for a read/write operation?
Isn't all heads moving as one single unit?

The heads move as a unit but when a head is aligned on a track on one
platter that does not mean that any of the other heads are aligned on the
corresponding tracks of the other platters. Each surface of each platter
has its own servo tracks.

Anyway how about two (or more) "head units" mounted inside a
hard-drive. Each unit taking care of each section of the platter. One
unit from center to middle and another for middle to the end. Also if
each unit was designed to have the possibility to move over the entire
platter it could have functioned as a "failsafe" in case one of the
heads was failing.
...but then again I guess someone else allready had this idea

Been done, repeatedly. Last one I recall was from Seagate.

Now if you want _fast_, you want fixed head disks. Nobody has made one of
those in ages.

 

[dannysdailys]

Not believing I'd clarify anything for Rod Speed, but, before h
attacks you, I'll explain what he probably meant

He didn't mean RAM on your system, he meant RAM in the drive. This i
now being done and hard drive speed closely matches RAM speed becaus
there is no hard drive per se

This method is also quite expensive, as you can imagine

But remember, no matter what you do to a hard drive, it's still slowe
then RAM by at least a thousand percent. Hard drives work i
milliseconds and RAM works in nanoseconds

I don't believe you'll see any hard drive speed strides that will b
significant. No physical hard drive will ever be able to compet
with an all solid state system and that's where the technology i
heading

Waxman wrote:[quote:6cbfee5f06] I can't image that *just* more ra
will give incredible speed. You will eventually end up waiting fo
the ram to be flushed to disk anyway.[/quote:6cbfee5f06

 

[Mike Redrobe]

Now if you want _fast_, you want fixed head disks. Nobody has made
one of those in ages.

Seek times of zero, but they were quite "big" drives:
http://cache.gizmodo.com/archives/images/new_hard_drive.jpg

 

[Waxhead]

Not believing I'd clarify anything for Rod Speed, but, before he
attacks you, I'll explain what he probably meant.

He didn't mean RAM on your system, he meant RAM in the drive. This is
now being done and hard drive speed closely matches RAM speed because
there is no hard drive per se.

This method is also quite expensive, as you can imagine.

But remember, no matter what you do to a hard drive, it's still slower
then RAM by at least a thousand percent. Hard drives work in
milliseconds and RAM works in nanoseconds.

I don't believe you'll see any hard drive speed strides that will be
significant. No physical hard drive will ever be able to compete
with an all solid state system and that's where the technology is
heading.

Waxman wrote:[quote:6cbfee5f06] I can't image that *just* more ram
will give incredible speed. You will eventually end up waiting for
the ram to be flushed to disk anyway.[/quote:6cbfee5f06]

FYI: I was aware that Rod was talking about the hard-drive cache I
guess I was a bit unclear on that...

 

[Rod Speed]

Rod Speed wrote

I can't image that *just* more ram will give incredible speed.

No one said anything about 'incredible speed'. Ram is a lot faster
than any physical drive can ever be, and thats what matters.

You will eventually end up waiting for
the ram to be flushed to disk anyway.

Nope, very few modern apps write huge amounts of data to a
drive now, so all you have to do is get rid of it to the platters
faster than you are capturing it. Thats true now with a single
normal hard drive, you dont even need faster than 7200 RPM
to capture 4 simultaneous digital TV streams to the hard drive.

Thanks for some excellent answers

No problem.

 

[Rod Speed]

In the old days when disk drives had a single dedicated servo track, it
was possible to do this.

It still is.

The cost of the additional electronics
made it cost prohibative for the most part.

The real reason you dont see dedicated servo surfaces
anymore is because that produces a lower storage
density, particularly with just 1 or 2 platters.

Nowadays, the cost of the additional electronics is much less.
The problem now is that to handle the higher track densities,
disk drives have embedded servos on each surface of each platter.
This allows the manufacturers to overcome differential expansion
and contraction between the heads but only one head can be brought
into alignment at a time.

Thats just plain wrong, most obviously with a single platter drive.

This restricts reading and writing to one head at a time.

No it doesnt.

 

[Rod Speed]

So what you are saying is that each head moves independently from the
others but only a single one can be used for a read/write operation?
Isn't all heads moving as one single unit?

Anyway how about two (or more) "head units" mounted inside a
hard-drive. Each unit taking care of each section of the platter.

That was seen with a few drives.

One unit from center to middle and another for middle to the end.

They dont need to cover separate zones if they
are on say opposite sides of the platter radially.

Also if each unit was designed to have the possibility
to move over the entire platter it could have functioned
as a "failsafe" in case one of the heads was failing.

Yes, but at a very significant extra cost.

...but then again I guess someone else allready had this idea

Yes, but it never really got used much.

 

[Rod Speed]

J. Clarke wrote

Seek times of zero, but they were quite "big" drives:
http://cache.gizmodo.com/archives/images/new_hard_drive.jpg

They didnt have to be that big, I had one that fitted into a standard
rack horizontally, a DEC RS09. Not that much storage capacity tho.

And not practical with modern track densitys anyway.

 

[Rod Speed]

Not believing I'd clarify anything for Rod Speed, but, before he
attacks you, I'll explain what he probably meant.

He didn't mean RAM on your system, he meant RAM in the drive. This
is now being done and hard drive speed closely matches RAM speed
because there is no hard drive per se.

This method is also quite expensive, as you can imagine.

But remember, no matter what you do to a hard drive, it's still
slower then RAM by at least a thousand percent. Hard drives work in
milliseconds and RAM works in nanoseconds.

I don't believe you'll see any hard drive speed strides that will be
significant. No physical hard drive will ever be able to compete
with an all solid state system and that's where the technology is
heading.

Waxman wrote:[quote:6cbfee5f06] I can't image that *just* more ram
will give incredible speed. You will eventually end up waiting for
the ram to be flushed to disk anyway.[/quote:6cbfee5f06]

FYI: I was aware that Rod was talking about the hard-drive cache

I wasnt actually, I was talking about replacing the platters and heads
with ram, so the drive has JUST ram. Those are buyable now.

 

[Arno Wagner]

Hi!

I just snooped around the net for information on how hard-drives work
(in detail). I found (if the information is up to date) that all the
read/write heads are moving as one single unit. Also I read that when
reading or writing the hard-drive usualy only use one head at the time.
Finaly I read that one side of each plattern is used for position
information.

So here's my idea. If the most reliable way to manufacture a read/write
heads is to make all heads move in unison why would you want to
read/write with only one head instead of distributing the data over all
heads. (just like a RAID array stripes data for example).

To overexplain this let's say you have a hard-drive with 8 platterns
and 8 read/write heads. If you write one byte (8bits) of data each bit
would be "assigned" it's own plattern making it possible to write one
byte with one "pulse". instead of "pulsing" this byte on one plattern.

I guess someone have allready had this idea but why isn't it smarter
for a hard-drive to internaly stripe the data over all platterns
instead of writing it one by one. Is it perhaps some reason who makes
it physically impossible for all read/write heads to do a read/write
all at once?

Does not work:
- The heads are not aligned that well, you can read/write only on
one at a time. There is no way around that.
In the long forgotten past, there actually was a separate
servo surface, where your idea would have had some merit,
but was obviously commercially unatractive.
- You would need to replicate r/w electronics for each head, making
it more expensive.
- The disk platters/heads are not the only thing that can fail.

Arno

 

[Folkert Rienstra]

Obviously never was, see next.

Nonsense, obviously, with all heads fixed together.

It's not. It's just a lot cheaper to build than independently movable heads.

Because most harddrive use just a few heads, dependent on their (size) capacity.

That's false even.
There's (on average) 3 platters in a harddrive using (on average) 6 heads.
Bigge(r)(st) harddrives use more platters, small ones use one and the smal-
lest even use less than that, using 1 head.


Current technology (very high densities) demands embedded servo on each
head. That demands all heads to move independently, which obviously they can't.

Because harddrives don't have a fixed number of platters, obviously.

Yes, in more ways than one.

Does not work:
- The heads are not aligned that well, you can read/write only on
one at a time. There is no way around that.

In the long forgotten past, there actually was a separate
servo surface, where your idea would have had some merit,

but was obviously commercially unatractive.

Wasn't sustainable too, technology wise.

- You would need to replicate r/w electronics for each head, making
it more expensive.

Not to mention to have 5 platters in each drive, independent of capacity.
Find a way to record far more data in a track (because track density is
finite with this type of servo system) or switch to independently movable
heads and embedded servo per head.

All harddrives will have the same size or different size harddrives will have
different speed characteristics. Since component count is the same for both
smaller and bigger sized harddrives they will basically cost the same to build.
None of that is very helpful in promoting this idea.

 

[Folkert Rienstra]

Seek times of zero,

but still suffering from rotational latency and head switch times.

 

[Curious George]

http://faq.storagereview.com/tiki-index.php?page=InternalRAID0
http://www.pcguide.com/ref/hdd/op/actMultiple-c.html