|
Home > Archive > Data Storage > September 2004 > new faster disk design idea
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
new faster disk design idea
|
|
|
| If you ever opened a hard drive, you will see a couple of platters and an
arm that has a head for each surface. So if 2 platters, then 4 surfaces and
4 heads. The arm moves the heads across the platters and the platters
rotate, so any spot on the disk can be seeked with both movement together.
So... why not have multiple heads on each surface? Imagine a row of heads
evenly spaced out on a bar at a right angle to the arm, the arm attaching at
the middle of the bar, forming a T shape. If there were 10 heads on each
bar, then you get 10x speed and 1/10 lateral seek time.
Good or bad idea? Feasible?
Clayton
| |
| Al Dykes 2004-06-26, 2:26 pm |
| In article <KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net>,
- C - <junky@ctny.org> wrote:
>If you ever opened a hard drive, you will see a couple of platters and an
>arm that has a head for each surface. So if 2 platters, then 4 surfaces and
>4 heads. The arm moves the heads across the platters and the platters
>rotate, so any spot on the disk can be seeked with both movement together.
>
>So... why not have multiple heads on each surface? Imagine a row of heads
>evenly spaced out on a bar at a right angle to the arm, the arm attaching at
>the middle of the bar, forming a T shape. If there were 10 heads on each
>bar, then you get 10x speed and 1/10 lateral seek time.
>
>Good or bad idea? Feasible?
>
Sort of.
You've obviously never seen a Univac Fastrand drum (64 heads visible
under plexiglass). 1968. $170k for 90MB with performance specs
almost as fast a modern CD reader.
http://www.fourmilab.ch/documents/univac/fastrand.html
IBM 2305 disk had a head per track ( a rousing 11MB, with zero seek
time.) (late 70's into the 80's. cost as much as a house.) I think
it needed water cooling.
Multiple heads per surface was a maintenance (or reliability)
nightmare and was only used at the bleading edge of what's possible,
which is definatly not mass market retail.
If you think about it, the huge storage arrays that corporations are
buying are the current incarnation of your idea; hundreds of heads
over spinning data, but each head has it's own spindle. WIth proper
I/O subsystem design this gives you the access times and transfer
rates you are looking for, See
http://www.emc.com/products/systems/DMX_series.jsp?
openfolder=storage_systems
Disk drives have gotten lots better, trust me.
--
Al Dykes
-----------
adykes at p a n i x . c o m
| |
| Brian Inglis 2004-06-26, 2:26 pm |
| On Tue, 15 Jun 2004 01:05:46 GMT in comp.arch.storage, "- C -"
<junky@ctny.org> wrote:
>If you ever opened a hard drive, you will see a couple of platters and an
>arm that has a head for each surface. So if 2 platters, then 4 surfaces and
>4 heads. The arm moves the heads across the platters and the platters
>rotate, so any spot on the disk can be seeked with both movement together.
>
>So... why not have multiple heads on each surface? Imagine a row of heads
>evenly spaced out on a bar at a right angle to the arm, the arm attaching at
>the middle of the bar, forming a T shape. If there were 10 heads on each
>bar, then you get 10x speed and 1/10 lateral seek time.
>
>Good or bad idea? Feasible?
IIRC it was done on the IBM 3380E (double density -- two heads) and
IBM 3380K (triple density -- three heads), same rotational speed and
same track-track and max seek times AFAIK, with the heads dangling
below the arms, for less than double or triple the not inconsiderable
price ($10/MB).
So what typically happened was drives were consolidated 2(/3) to 1,
saving money, space, power etc., but that increased the number of
files and I/Os per drive by 2(/3), and also reduced the number of
simultaneous I/Os, seeks, and transfers by 2(/3), resulting in worse
overall performance.
If the drives had been addressed as if they had more tracks/cylinder
instead of more cylinders, and used for only sequentially accessed
files, users might have seen benefits.
More than three heads per were probably tried but may have had
problems with arm mass, flutter, head-track registration, flying
height, or various other physical considerations you can probably
think of.
--
Thanks. Take care, Brian Inglis Calgary, Alberta, Canada
Brian.Inglis@CSi.com (Brian dot Inglis at SystematicSw dot ab dot ca)
fake address use address above to reply
| |
| Peter da Silva 2004-06-26, 2:26 pm |
| In article <KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net>,
- C - <junky@ctny.org> wrote:
> So... why not have multiple heads on each surface?
Whoa, what's next, 8-track? Disco? The Nixon presidency over again?
This was common in the '70s, but the heavier the head is the slower it can
move and the more complicated *stuff* you have on the head the heavier it
is and the less reliable it is. It's an idea whose time has gone...
--
I've seen things you people can't imagine. Chimneysweeps on fire over the roofs
of London. I've watched kite-strings glitter in the sun at Hyde Park Gate. All
these things will be lost in time, like chalk-paintings in the rain. `-_-'
Time for your nap. | Peter da Silva | Har du kramat din varg, idag? 'U`
| |
| Adrian 'Dagurashibanipal' von Bidder 2004-06-26, 2:26 pm |
| -----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Clinging to sanity, - C - mumbled in his beard:
[more heads on platters]
Yo!
Didn't HP have a prototype that went in that direction? Only they did not
have just more heads, but duplicated the whole electronics/heads/arms part,
essentially 2 independent drives accessing the same spindle. The aim was to
increase number of transactions/time, not bandwidth.
The name kittyhawk bounces around in my head, but I can't find anything in
google right now.
cheers
- -- vbi
- --
Baker's First Law of Federal Geometry:
A block grant is a solid mass of money surrounded on all sides by
governors.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (GNU/Linux)
Comment: get my key from http://fortytwo.ch/gpg/92082481
iKcEARECAGcFAkDPKxtgGmh0dHA6Ly9mb3J0eXR3
by5jaC9sZWdhbC9ncGcvZW1h
aWwuMjAwMjA4MjI/ dmVyc2lvbj0xLjUmbWQ1c3VtPTVkZmY4NjhkMTE4
NDMyNzYw
NzFiMjVlYjcwMDZkYTNlAAoJEIukMYvlp/fWtDEAoO/i5V4MPnXjaoRt/0F0m+Sc
ide+AJ9afnn7J75103KtzGLeAdKi7GtgRw==
=N1sY
-----END PGP SIGNATURE-----
| |
| Ron Reaugh 2004-06-26, 2:26 pm |
| Such a SCSI drive was commercially shipped about 5 years ago....the Seagate
ST12450W.
"- C -" <junky@ctny.org> wrote in message
news:KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net...
> If you ever opened a hard drive, you will see a couple of platters and an
> arm that has a head for each surface. So if 2 platters, then 4 surfaces
and
> 4 heads. The arm moves the heads across the platters and the platters
> rotate, so any spot on the disk can be seeked with both movement together.
>
> So... why not have multiple heads on each surface? Imagine a row of heads
> evenly spaced out on a bar at a right angle to the arm, the arm attaching
at
> the middle of the bar, forming a T shape. If there were 10 heads on each
> bar, then you get 10x speed and 1/10 lateral seek time.
>
> Good or bad idea? Feasible?
Not feasible generally. Only a single head per physically actuator can be
active at any given time as the track servo following can only follow one
master at a time. The ST12450W had two complete head/actuator systems.
| |
| Bill Todd 2004-06-26, 2:26 pm |
|
"- C -" <junky@ctny.org> wrote in message
news:KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net...
> If you ever opened a hard drive, you will see a couple of platters and an
> arm that has a head for each surface. So if 2 platters, then 4 surfaces
and
> 4 heads. The arm moves the heads across the platters and the platters
> rotate, so any spot on the disk can be seeked with both movement together.
>
> So... why not have multiple heads on each surface?
Because, as others have noted, it increases the mass and inertia of the
assembly. When you get twice the capacity as a result (as you do with a
head on each side of a platter, using a single arm), it's worth it;
otherwise, it isn't.
Imagine a row of heads
> evenly spaced out on a bar at a right angle to the arm, the arm attaching
at
> the middle of the bar, forming a T shape. If there were 10 heads on each
> bar, then you get 10x speed and 1/10 lateral seek time.
No, you don't - not anything like it. In fact, performance may well get
worse.
The only way to significantly decrease the movement required for an average
seek is to make that bar nearly as wide as the disk surface (otherwise, you
still have to move it almost as far to get to a track 1/3 of the surface
away). That makes the bar *much* more massive, so it takes longer both to
move it a given distance and to stabilize it once it reaches its
destination.
The ultimate such 'bar' has already been described: it has a head for each
track on the surface, and doesn't move at all - and in fact was therefore a
performance win (though at significant cost). That was feasible 'way back
when, but at current track densities there's nowhere nearly enough room for
a head per track on such a bar.
Using multiple arms to access a single platter has also been mentioned.
However, you can get nearly the same effect simply by using two conventional
disks (plus get twice the capacity in the process) - and the production
efficiencies of using standard units make this a cheaper solution than a
single, low-volume dual-armed disk.
- bill
| |
|
| Ron Reaugh wrote:
> Such a SCSI drive was commercially shipped about 5 years ago....the Seagate
> ST12450W.
>
But the heads operating in parallel were very close to each other.
If that were different, thermal expansion would mean one of the heads
would mistrack.
Thomas
| |
| Robert Wessel 2004-06-26, 2:26 pm |
| "Bill Todd" <billtodd@metrocast.net> wrote in message news:<4IOdnQ_hl93QOEzdRVn-vA@metrocast.net>...
> "- C -" <junky@ctny.org> wrote in message
> news:KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net...
> and
>
> Because, as others have noted, it increases the mass and inertia of the
> assembly. When you get twice the capacity as a result (as you do with a
> head on each side of a platter, using a single arm), it's worth it;
> otherwise, it isn't.
>
> Imagine a row of heads
> at
>
> No, you don't - not anything like it. In fact, performance may well get
> worse.
>
> The only way to significantly decrease the movement required for an average
> seek is to make that bar nearly as wide as the disk surface (otherwise, you
> still have to move it almost as far to get to a track 1/3 of the surface
> away). That makes the bar *much* more massive, so it takes longer both to
> move it a given distance and to stabilize it once it reaches its
> destination.
Not to mention that you're only optimizing away the fastest part of
the seek. Head acceleration, deceleration and settling time and
whatnot are all still there for the nominally shorter seeks, and, as
you mentioned, will probably get slower due to increased arm mass.
And given that most seeks are quite short anyway, you'd need a lot of
heads to make this even marginally effective.
| |
|
| GREAT IDEA! WHY DON'T YOU GO INTO THE RAZOR BUSINESS AND MAKE A RAZOR
WITH 18 BLADES.
"- C -" <junky@ctny.org> wrote in message
news:KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net...
> If you ever opened a hard drive, you will see a couple of platters and an
> arm that has a head for each surface. So if 2 platters, then 4 surfaces
and
> 4 heads. The arm moves the heads across the platters and the platters
> rotate, so any spot on the disk can be seeked with both movement together.
>
> So... why not have multiple heads on each surface? Imagine a row of heads
> evenly spaced out on a bar at a right angle to the arm, the arm attaching
at
> the middle of the bar, forming a T shape. If there were 10 heads on each
> bar, then you get 10x speed and 1/10 lateral seek time.
>
> Good or bad idea? Feasible?
>
> Clayton
>
>
| |
| Mark Landin 2004-06-26, 2:26 pm |
| On Fri, 18 Jun 2004 02:08:31 -0600, "dafon" <dafon@sdafon.net> wrote:
> GREAT IDEA! WHY DON'T YOU GO INTO THE RAZOR BUSINESS AND MAKE A RAZOR
>WITH 18 BLADES.
No need to be derogatory. It was a legitimate question.
| |
|
| True. No need to be derogatory.
It just felt good!
"Mark Landin" <mark.landin@tdwilliamson.com> wrote in message
news:c0v5d0ht357k7c7m0rcoo1oa40gcl9jbe7@
4ax.com...
> On Fri, 18 Jun 2004 02:08:31 -0600, "dafon" <dafon@sdafon.net> wrote:
>
RAZOR[vbcol=seagreen]
>
> No need to be derogatory. It was a legitimate question.
>
>
| |
| Akheel Ahmed 2004-06-28, 7:22 pm |
| Hi,
Actually its not a bad idea to have 2 heads attached to one arm in such
a way that if one is over the outermost cylinder the other is over the
innermost cylinder(ps see fig.). by doing this, you cut down both the
seek time and the rotational latency by approx 50%.This can be a
feasible first step towards faster disks.
Spindle
|
|
___ V ___
Head 1 --> |___\ . /___| <-- Head 2
\ /
\ /
\ /
\ /
\______/
||
||
||
|-----|
| | <-- Actuator
|_____|
Akheel
dafon wrote:
> GREAT IDEA! WHY DON'T YOU GO INTO THE RAZOR BUSINESS AND MAKE A RAZOR
> WITH 18 BLADES.
>
>
> "- C -" <junky@ctny.org> wrote in message
> news:KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net...
>
>
> and
>
>
> at
>
>
>
>
| |
| Al Dykes 2004-06-28, 7:22 pm |
|
Did you read the thread for this topic before you posted ? I guess not.
In article <40E034D9.4090807@njit.edu>, Akheel Ahmed <aa97@njit.edu> wrote:
>Hi,
>
>Actually its not a bad idea to have 2 heads attached to one arm in such
>a way that if one is over the outermost cylinder the other is over the
>innermost cylinder(ps see fig.). by doing this, you cut down both the
>seek time and the rotational latency by approx 50%.This can be a
>feasible first step towards faster disks.
>
>
>
> Spindle
> |
> |
> ___ V ___
> Head 1 --> |___\ . /___| <-- Head 2
> \ /
> \ /
> \ /
> \ /
> \______/
> ||
> ||
> ||
> |-----|
> | | <-- Actuator
> |_____|
>
>
>
>
>Akheel
>
>
>
>dafon wrote:
>
--
Al Dykes
-----------
adykes at p a n i x . c o m
| |
| Ron Reaugh 2004-06-28, 7:22 pm |
|
"Akheel Ahmed" <aa97@njit.edu> wrote in message
news:40E034D9.4090807@njit.edu...
> Hi,
>
> Actually its not a bad idea to have 2 heads attached to one arm in such
> a way that if one is over the outermost cylinder the other is over the
> innermost cylinder(ps see fig.). by doing this, you cut down both the
> seek time and the rotational latency by approx 50%.This can be a
> feasible first step towards faster disks.
Nope, doesn't work. Read up on servo track following.
| |
| danco@ns2.pebble.org 2004-09-23, 5:46 pm |
| In article <KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net>,
- C - wrote:
> So... why not have multiple heads on each surface? Imagine a row of heads
> evenly spaced out on a bar at a right angle to the arm, the arm attaching at
> the middle of the bar, forming a T shape. If there were 10 heads on each
> bar, then you get 10x speed and 1/10 lateral seek time.
Blue sky time:
Perhaps we can eliminate the arm altogether and instead place an imovable
bar shaped IC above and perpendicular to the platter. The IC has
thousands upon thousands of little read and write heads etched onto its
surface to cover a small swath of the platter from the inside edge to the
outside edge. No head movement. No settling time. Zero track to track
seek time. Just rotational latency and head switch time. If reading
and writing to from multiple parallel heads at the same time is possible
using this non-moving set of heads, then head switch time can be reduced.
Perhaps rotational latency can even be reduced by placing multiples of
these bar ICs above the platter and using them simultaniously.
Feasible today? Feasible tomorrow? Or just an impractical bunch of BS
both now and forever?
I think likely something totally different (and technologically much less
Rube Goldberg oriented) will come along to make the whole rotating platter
storage media concept obsolete -- long before stuff like the above
ever becomes practical.
- Dan
| |
| danco@ns2.pebble.org 2004-09-23, 8:45 pm |
| In article <PF%Dc.39477$OB3.32657@bgtnsc05-news.ops.worldnet.att.net>,
Ron Reaugh wrote:
> Nope, doesn't work. Read up on servo track following.
Perhaps if you can have thousands and thousands of little heads on an
imovable IC that floats above the platter and covers the platter from
the inside to the outside edge (in the shape of a bar), with each head
having a width on the order of a transistor etched onto the surface of
the IC, you can do your servo track following by switching between
adjacent heads as the track "wanders." Likely impractical with today's
technology, eh?
- Dan
| |
| Ron Reaugh 2004-09-23, 8:45 pm |
|
<danco@ns2.pebble.org> wrote in message
news:slrncl6jeo.sqj.danco@ns2.pebble.org...
> In article <KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net>,
> - C - wrote:
>
heads[vbcol=seagreen]
attaching at[vbcol=seagreen]
>
> Blue sky time:
>
> Perhaps we can eliminate the arm altogether and instead place an imovable
> bar shaped IC above and perpendicular to the platter. The IC has
> thousands upon thousands of little read and write heads etched onto its
> surface to cover a small swath of the platter from the inside edge to the
> outside edge. No head movement. No settling time. Zero track to track
> seek time. Just rotational latency and head switch time. If reading
> and writing to from multiple parallel heads at the same time is possible
> using this non-moving set of heads, then head switch time can be reduced.
> Perhaps rotational latency can even be reduced by placing multiples of
> these bar ICs above the platter and using them simultaniously.
>
> Feasible today? Feasible tomorrow? Or just an impractical bunch of BS
> both now and forever?
>
> I think likely something totally different (and technologically much less
> Rube Goldberg oriented) will come along to make the whole rotating platter
> storage media concept obsolete -- long before stuff like the above
> ever becomes practical.
If you spin the disks backwards then negative polarity magnetics can be read
faster.
| |
| Tim Boyer 2004-09-24, 2:46 am |
| On Thu, 23 Sep 2004 23:31:52 GMT, danco@ns2.pebble.org wrote:
>In article <KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net>,
>- C - wrote:
>
>
>Blue sky time:
>
>Perhaps we can eliminate the arm altogether and instead place an imovable
>bar shaped IC above and perpendicular to the platter. The IC has
>thousands upon thousands of little read and write heads etched onto its
>surface to cover a small swath of the platter from the inside edge to the
>outside edge. No head movement. No settling time. Zero track to track
>seek time. Just rotational latency and head switch time. If reading
>and writing to from multiple parallel heads at the same time is possible
>using this non-moving set of heads, then head switch time can be reduced.
>Perhaps rotational latency can even be reduced by placing multiples of
>these bar ICs above the platter and using them simultaniously.
>
>Feasible today? Feasible tomorrow? Or just an impractical bunch of BS
>both now and forever?
>
<chuckle> Feasible 30 years ago? I ran one on a Data General mini back in the
late 70s. Of course, the storage capacity was a bit smaller then. They were
called 'fixed-head' or 'head-per-track' disks. Damned expensive, as I recall.
--
tim boyer
tim@denmantire.com
| |
| Brian Inglis 2004-09-24, 2:46 am |
| ffOn Thu, 23 Sep 2004 23:31:52 GMT in comp.arch.storage,
danco@ns2.pebble.org wrote:
>In article <KXrzc.27433$Yd3.14176@newsread3.news.atl.earthlink.net>,
>- C - wrote:
>
Greater head mass slows the arm and takes longer to settle.
[vbcol=seagreen]
>Perhaps we can eliminate the arm altogether and instead place an imovable
>bar shaped IC above and perpendicular to the platter. The IC has
>thousands upon thousands of little read and write heads etched onto its
>surface to cover a small swath of the platter from the inside edge to the
>outside edge. No head movement. No settling time. Zero track to track
If they could make heads so small, they would have done so already,
and seek time would be reduced somewhat proportional to the mass.
>seek time. Just rotational latency and head switch time. If reading
Rotational latency is the current bottleneck, as heavier platters spin
slower and lighter platters shatter due to rotational stress. Current
need is for lighter, stronger, cheap substrates that can spin faster,
and have magnetic media permanently bonded to them.
--
Thanks. Take care, Brian Inglis Calgary, Alberta, Canada
Brian.Inglis@CSi.com (Brian[dot]Inglis{at}SystematicSW[dot]ab[dot]ca)
fake address use address above to reply
| |
| Jesper Monsted 2004-09-24, 5:45 pm |
| danco@ns2.pebble.org wrote in news:slrncl6jeo.sqj.danco@ns2.pebble.org:
> Perhaps we can eliminate the arm altogether and instead place an
> imovable bar shaped IC above and perpendicular to the platter. The IC
> has thousands upon thousands of little read and write heads etched
> onto its surface to cover a small swath of the platter from the inside
> edge to the outside edge. No head movement. No settling time. Zero
> track to track seek time. Just rotational latency and head switch
> time. If reading and writing to from multiple parallel heads at the
> same time is possible using this non-moving set of heads, then head
> switch time can be reduced. Perhaps rotational latency can even be
> reduced by placing multiples of these bar ICs above the platter and
> using them simultaniously.
You'd spend more on the r/w-heads than you would by just buying solid state
gear, i'd wager.
--
/Jesper Monsted
|
|
|
|
|