http://www.geek.com/news/geeknews/20...1212023056.htm

An interesting user comment:

At some level... it really is amazing (4:59pm EST Fri Dec 12 2003)
[Reminiscing, pardons...]

I mean - it was only back in the 1930's for history's sake, that the
first steel-wire recorders were developed, and in the 1940's came "tape"
recorders, which sported flakey films of ground-up 'burned' hematite (a
modestly magnetic iron ore rock) on a nitrocellulose substrate. For the
longest time, the ground-rock film was king... the first true spinning
digital disks used it, and it even made it as far as the 30 megabyte
5.25 inch "Seagate" drives. But those lil' particles had to go.

Too noisy, not enough energy stored per unit area, way too large, and
damned difficult to set down in smooth layers. The industry transitioned
to metal-particle slurries, which sufficed for awhile, then gradually
went to directly "plated" aluminum platters. The plating was actually a
high voltage sputtering operation, and the media had to be passivated
with dilute nitric then sulfuric acids and a number of interesting
"washes" of ammonia and hexabutyl hexanoate. Anyway...

The issue though - for the longest time - was that the "heads" were made
from itty bitty "U's" of a magnetic flux core that was literally wrapped
with several turns (!!! can you imagine) of fine wire, to constitute the
read/write head. The 'gap' was horizontal, so the smaller the gap, the
less the write-field would leak out, making it harder and harder to
impinge 'bits' on the underlying spinning media.

Then researchers discovered some really amazing compounds that had the
property of changing their resistance in response to fairly modest
changes in local magnetic field. These made for extremely sensitive READ
heads, which could successfuly read bits far smaller than the old "U"
heads. But how to write these darn bits? Well... then the idea of a
'monopolar' head took over: a single tiny pole, instead of a pair, and
relying on Maxwell's law of magnetics (which guarantees that all paths
contain a net of 'zero' magnetic field, once integrated). So, the other
pole could be big, and essentially virtual. Bits were recorded
vertically, and things immediately got much better.

These magnetoresistive heads and monopole bit-writers propelled disk
capacity (which had more or less begun to stagnate at the 1.2 and 2.4
gigabyte 5.25 inch, full height monstrosity) to rapidly moving new and
much lower priced points. Today CMR ("colossal magneto-resistive")
heads, and nanometer-scale write-heads are powering the 70 gigabit per
sq. inch. densities now cheaply available.

The concept of nanopatterning (which may well prove to be an absolutely
excellent use for the 'nanoprinting' impression technology spoken of
elsewhere) will make sure magnetic domains are at least well formed,
uniform in size, uniform in energy-capacity, and uniform in their
resistance to change (data loss). The superparamagnetic limit is
sidestepped quite nicely when each domain is physically separated from
each other domain by a gap. Hard for neighboring magnetic fields to flip
OUR field, if we're each an island unto ourselves.

So, the bit densities will rise to the 100 gigaBYTE to maybe 2-3 times
higher limit. As another poster on a different geek.com forum pointed
out, what the hell good is all those terabytes, if it takes months to
perform a defrag?

In essence, I/O is falling so far behind processing speed (literally -
in 1981 a 5 inch hard drive delivered 2 megabyte per second performance,
and the CPUs of the day (8086) could only slog around 10 megabytes per
second on their memory busses themselves... Today, the
highest-of-the-highest speed hard drives can sustain 70 megabytes per
second throughput, but CPUs are able to slush around 3500 megbyte or
more per second on their busses.)

The change is going to have to be a bit expensive, and pretty heavily
leveraged off of what microelectronics CAN do: pipeline everything. I
see the heads being redone to have 16 or 32 (or 36 for ECC) complete
read islands, and write islands. I see the supporting electronics able
to read these and cross-correlate the signals into a bit stream that is
20 to 30 times the throughput of todays drives. I see 64 megabyte (or
more!) on-drive caches, to further speed up operations, and smart
statistical "look-ahead" circuitry to pre-position and pre-read data for
delivery to the CPU. I also see the need for substantially faster BUS
interfaces - 1 GByte/sec at a minimum, and more like 4
GByte/sec/64-bit/500MHz to really keep the data flowing.

Then... and maybe then... will the hard drives be fast enough for us to
have nearly "instant booting", and so on. Data needs to get both ON and
OFF those disks as fast as possible.

- by GoatGuy

"John H." wrote:
>
> http://www.geek.com/news/geeknews/20...1212023056.htm
>
> An interesting user comment:
>
> At some level... it really is amazing (4:59pm EST Fri Dec 12 2003)
> [Reminiscing, pardons...]
>
> I mean - it was only back in the 1930's for history's sake, that the
> first steel-wire recorders were developed, and in the 1940's came "tape"
> recorders, which sported flakey films of ground-up 'burned' hematite (a
> modestly magnetic iron ore rock) on a nitrocellulose substrate. For the
> longest time, the ground-rock film was king... the first true spinning
> digital disks used it, and it even made it as far as the 30 megabyte
> 5.25 inch "Seagate" drives. But those lil' particles had to go.

Hello, John:

"Steel-wire recorders" were originally patented (in Denmark) in 1898,
and by 1915 had become essentially "perfected."

"Nitrocellulose substrate" was a post-WWII American improvement over
paper magnetic tape, a German innovation of the 1930's and '40's.

> Too noisy, not enough energy stored per unit area, way too large, and
> damned difficult to set down in smooth layers. The industry transitioned
> to metal-particle slurries, which sufficed for awhile, then gradually
> went to directly "plated" aluminum platters. The plating was actually a
> high voltage sputtering operation, and the media had to be passivated
> with dilute nitric then sulfuric acids and a number of interesting
> "washes" of ammonia and hexabutyl hexanoate. Anyway...
>
> The issue though - for the longest time - was that the "heads" were made
> from itty bitty "U's" of a magnetic flux core that was literally wrapped
> with several turns (!!! can you imagine) of fine wire, to constitute the
> read/write head. The 'gap' was horizontal, so the smaller the gap, the
> less the write-field would leak out, making it harder and harder to
> impinge 'bits' on the underlying spinning media.

Hard disks drives were invented by IBM, during the 1950's. This company
has lead the way in HDD technology, ever since.

<large cut, for brevity>

> In essence, I/O is falling so far behind processing speed (literally -
> in 1981 a 5 inch hard drive delivered 2 megabyte per second performance,
> and the CPUs of the day (8086) could only slog around 10 megabytes per
> second on their memory busses themselves... Today, the
> highest-of-the-highest speed hard drives can sustain 70 megabytes per
> second throughput, but CPUs are able to slush around 3500 megbyte or
> more per second on their busses.)
>
> The change is going to have to be a bit expensive, and pretty heavily
> leveraged off of what microelectronics CAN do: pipeline everything. I
> see the heads being redone to have 16 or 32 (or 36 for ECC) complete
> read islands, and write islands. I see the supporting electronics able
> to read these and cross-correlate the signals into a bit stream that is
> 20 to 30 times the throughput of todays drives. I see 64 megabyte (or
> more!) on-drive caches, to further speed up operations, and smart
> statistical "look-ahead" circuitry to pre-position and pre-read data for
> delivery to the CPU. I also see the need for substantially faster BUS
> interfaces - 1 GByte/sec at a minimum, and more like 4
> GByte/sec/64-bit/500MHz to really keep the data flowing.
>
> Then... and maybe then... will the hard drives be fast enough for us to
> have nearly "instant booting", and so on. Data needs to get both ON and
> OFF those disks as fast as possible.
>
> - by GoatGuy

No, the ultimate goal is reasonably-priced, solid state devices; only
these would have even remote hopes of keeping pace with modern CPU's, it
appears. I fear that inherent mechanical limitations will eventually
doom conventional HDD's.


Cordially,
John Turco <(E-Mail Removed)>

I don't know which 5" drive produced 2MB/s in '81. Must have been "server
class" drive. ST412 interface had 5 Mb/s (MFM), or 7.5 Mb/s (RLL, though the
interface was called something different, with R suffix). Interleave was
also required, which made the STR even less.

In current drives, 70 MB/s is top of the line IDE drive speed. I suppose,
enterprise drives give even more.

"John H." <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...
> http://www.geek.com/news/geeknews/20...1212023056.htm
>
>
> In essence, I/O is falling so far behind processing speed (literally -
> in 1981 a 5 inch hard drive delivered 2 megabyte per second performance,
> and the CPUs of the day (8086) could only slog around 10 megabytes per
> second on their memory busses themselves... Today, the
> highest-of-the-highest speed hard drives can sustain 70 megabytes per
> second throughput, but CPUs are able to slush around 3500 megbyte or
> more per second on their busses.)
>

John Turco <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...
> John H wrote

>> Then... and maybe then... will the hard drives be
>> fast enough for us to have nearly "instant booting",

Completely routine to have that now if you want it.

>> and so on. Data needs to get both ON
>> and OFF those disks as fast as possible.

Not necessarily.

> No, the ultimate goal is reasonably-priced, solid state devices;

It remains to be seen if those will ever replace
other media, particularly for longer term storage.

> only these would have even remote hopes of
> keeping pace with modern CPU's, it appears.

Thats easily fixed by getting it off the other
media well before the cpu needs to use it.

> I fear that inherent mechanical limitations
> will eventually doom conventional HDD's.

We'll see. Plenty have been claiming that for a long time.

Even if GoatGuy has a historical fact or two wrong, it's still an
interesting post. HD I/O speeds *are* falling behind relative to CPU
speeds.

> No, the ultimate goal is reasonably-priced, solid state devices; only
> these would have even remote hopes of keeping pace with modern CPU's, it
> appears. I fear that inherent mechanical limitations will eventually
> doom conventional HDD's.

I hope so. Wouldn't it be great if you could buy a PCI card today for
<$200 with 4GB solid state storage (enough for the boot partition).
Just think, zero access time, 133MB transfer rates for all files no
matter how fragmented. Or better yet, put the chip(s) on high-end MBs
for much higher transfer rates.

But for spun storage, adding more heads per surface like GoatGuy
suggested and getting the transfer rate up...way up...sounds good to me.
The only things HD manufacturers seem to care about is price and storage
capacity - not performance.

"Alexander Grigoriev" <(E-Mail Removed)> wrote in message news:vKaDb.4817$(E-Mail Removed)
> I don't know which 5" drive produced 2MB/s in '81. Must have been "server
> class" drive. ST412 interface had 5 Mb/s (MFM), or 7.5 Mb/s (RLL, though the
> interface was called something different, with R suffix). Interleave was
> also required, which made the STR even less.
>
> In current drives, 70 MB/s is top of the line IDE drive speed. I suppose,
> enterprise drives give even more.

What enterprise drives.

>
> "John H." <(E-Mail Removed)> wrote in message news:(E-Mail Removed)...
> > http://www.geek.com/news/geeknews/20...1212023056.htm
> >
> >
> > In essence, I/O is falling so far behind processing speed (literally -
> > in 1981 a 5 inch hard drive delivered 2 megabyte per second performance,
> > and the CPUs of the day (8086) could only slog around 10 megabytes per
> > second on their memory busses themselves... Today, the
> > highest-of-the-highest speed hard drives can sustain 70 megabytes per
> > second throughput, but CPUs are able to slush around 3500 megbyte or
> > more per second on their busses.)

John H <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...

> Even if GoatGuy has a historical fact or two wrong,
> it's still an interesting post. HD I/O speeds *are*
> falling behind relative to CPU speeds.

Sure, but no PC has ever used the HD I/O directly to the cpu anyway.

>> No, the ultimate goal is reasonably-priced, solid state devices;
>> only these would have even remote hopes of keeping pace with
>> modern CPU's, it appears. I fear that inherent mechanical
>> limitations will eventually doom conventional HDD's.

> I hope so. Wouldn't it be great if you could buy a PCI card today for
> <$200 with 4GB solid state storage (enough for the boot partition).

You can, its just on the motherboard instead of
an addon PCI card for a bit more money than that.

Hardly anyone actually needs that much that fast.

Plenty of servers do it like that tho.

> Just think, zero access time,

Thats what any cache does.

> 133MB transfer rates for all files no matter how fragmented.

See above with servers.

> Or better yet, put the chip(s) on high-end
> MBs for much higher transfer rates.

Buyable now.

> But for spun storage, adding more heads per
> surface like GoatGuy suggested and getting the
> transfer rate up...way up... sounds good to me.

Pointless if you have the cache sized properly.

> The only things HD manufacturers seem to care about
> is price and storage capacity - not performance.

Oh bullshit. The bleeding edge has always been about performance.

On Tue, 16 Dec 2003 09:22:05 +1100, "Rod Speed" <(E-Mail Removed)>
wrote:

> > Wouldn't it be great if you could buy a PCI card today for
> > <$200 with 4GB solid state storage (enough for the boot partition).
>
> You can, its just on the motherboard instead of
> an addon PCI card for a bit more money than that.

What product are you talking about? I don't believe I've seen it.
>
> Hardly anyone actually needs that much that fast.
>
> Plenty of servers do it like that tho.
>
> > Just think, zero access time,
>
> Thats what any cache does.
>
Not 4GBs (on a home system).

John H <(E-Mail Removed)> wrote in message
news:(E-Mail Removed)...
> Rod Speed <(E-Mail Removed)> wrote

>> John H <(E-Mail Removed)> wrote

>>> Wouldn't it be great if you could buy a PCI card today for <$200
>>> with 4GB solid state storage (enough for the boot partition).

>> You can, its just on the motherboard instead of
>> an addon PCI card for a bit more money than that.

> What product are you talking about?

A motherboard that allows that much ram to
be installed. There's quite a few that do now.

> I don't believe I've seen it.

You want to get out more.

There isnt much point with personal desktop systems
but plenty of servers have total ram in that class now.

>> Hardly anyone actually needs that much that fast.

>> Plenty of servers do it like that tho.

>>> Just think, zero access time,

>> Thats what any cache does.

> Not 4GBs (on a home system).

You'd be surprised. Quite a few motherboards used in
personal desktop systems can have that much ram now.

Like I said, no real point in doing it, but you can if you want to.

On Mon, 15 Dec 2003 21:27:15 GMT
John H. <(E-Mail Removed)> wrote:

> Even if GoatGuy has a historical fact or two wrong, it's still an
> interesting post. HD I/O speeds *are* falling behind relative to CPU
> speeds.
>
> > No, the ultimate goal is reasonably-priced, solid state devices;
> > only these would have even remote hopes of keeping pace with modern
> > CPU's, it appears. I fear that inherent mechanical limitations will
> > eventually doom conventional HDD's.
>
> I hope so. Wouldn't it be great if you could buy a PCI card today for
> <$200 with 4GB solid state storage (enough for the boot partition).
> Just think, zero access time, 133MB transfer rates for all files no
> matter how fragmented. Or better yet, put the chip(s) on high-end MBs
> for much higher transfer rates.
>
> But for spun storage, adding more heads per surface like GoatGuy
> suggested and getting the transfer rate up...way up...sounds good to
> me. The only things HD manufacturers seem to care about is price and
> storage capacity - not performance.

Do not assume that just because something is solid state it is fast.
Check the comparisons between CF memory and Microdrive and you'll find
that in that case the disk beats the solid state memory. Cheap, fast,
large solid state memory is going to take a while.

--
--
--John
Reply to jclarke at ae tee tee global dot net
(was jclarke at eye bee em dot net)