Why don't HD-designers use their heads?
Laurens Leurs
productfamily, eg Maxtor DiamondMax 2880, the 1 platter disk is only
marginally slower that the 2 or 3 platter harddisk.
Why don't they design the harddisk so it writes with all heads
simultaneously on multiplatter HDs? That way performance would double on
a 2 platter drive or triple on a 3 platter drive.
Just wondering.......
remove no-spam text */ibm.net
lle...@mail.dma.be (Laurens Leurs) said:
>Just wondering.......
re-inventing striping? But it may be rather difficult to keep data on two
or three platters synchronised when read/write heads are heating up (maybe
in a different way) and expanding.
hans.boer@/* remove this text before replying */ibm.net
"The Roman Empire failed because it was not Year '0' compliant"
-----------------------------------------------------------
KALLE
wrote:
>If you look at the performance of different harddisks within the same
>productfamily, eg Maxtor DiamondMax 2880, the 1 platter disk is only
>marginally slower that the 2 or 3 platter harddisk.
Theroreticaly they must be marginal faster, because fewer heads are
lighter and so can be faster moved.
>
>Why don't they design the harddisk so it writes with all heads
>simultaneously on multiplatter HDs? That way performance would double on
>a 2 platter drive or triple on a 3 platter drive.
>
>Just wondering.......
You have to have all the write read amplifiers etc multiple times,
maybe there are problems of keeping all heads at the right position
(remeber the racks are very very tiny!)
You would get a better burstspeed, but seek time and latency would
stay the same, so you won't get so much profit out of it, as you
expect.
I've asked the Question to me bevore, too, and I came to the answer:
They surly know a reason why they don't do!
Just my 2 Cents
KALLE
see http://www.jump.net/~lcs/kalle/index.htm for ASUS P55T2P4 and other information.
(-: these humor brackets mean this part is meant to be funny, not to offend you :-)
Rusty Wright
much of the time. Also, once it gets to the correct track it has to
wait for the correct sector to rotate under the head; another very
slow process. Once the correct sector arrives under the head and it
starts reading or writing that's the fast part.
That's why the 7200 rpm disks are faster; takes less time for the
correct sector to get to the head once it's on the correct track.
A trick that they could use that's been used before is to have
multiple heads for each platter. E.g., have 2 heads for each platter,
then it has to seek only half as far. In other words, the first head
is for the first half of the platter and the second head is for the
rest of the platter.
Ron Reaugh
Laurens Leurs <lle...@mail.dma.be> wrote in article
<1dc7gn9.3vedrx1oavjq0N@[193.121.242.209]>...
> If you look at the performance of different harddisks within the same
> productfamily, eg Maxtor DiamondMax 2880, the 1 platter disk is only
> marginally slower that the 2 or 3 platter harddisk.
>
> Why don't they design the harddisk so it writes with all heads
> simultaneously on multiplatter HDs?
That's easy. The tolerances are so tight and the tracks are so close
together and narrow that mechanical aging and thermal expansion moves the
heads and tracks all over the place. The onboard electronics contain servo
technology whereby the heads actively find and follow where the tracks is
in real rotational time. The process only works on a single track at a
time. To have multiple heads active simultaneously then each head would
have to have its own independent actuator mechanism and R/W electronics
which would make a HD VERY expensive. The last drive to do that was
discontinued 3 years ago(ST12450W).
pit...@my-dejanews.com
>productfamily, eg Maxtor DiamondMax 2880, the 1 platter disk is only
>marginally slower that the 2 or 3 platter harddisk.
This is true because the drive must spend a greater portion of it's
overall time moving the head from track to track versus just moving from
platter to platter (a head switch can be accomplished in the
microprocessor/DSP/amplifier system of a HDD, a track-to-track seek
requires a physical actuator movement and the associated recalibration).
>Why don't they design the harddisk so it writes with all heads
>simultaneously on multiplatter HDs? That way performance would double on
Three reasons:
a) Thermal expansion. Since data is packed onto the drive so
tightly, even a degree of operating temperature difference between the top
of the drive assembly versus the bottom of the HDA assembly means that
everything is thrown out of whack. Since physical actuator movements are
probably required for recalibrating the drive in the event of hitting a
wrong track, it would be impossible to have multiple heads on the same
actuator active at the same time. This perhaps would be acceptable
and would work somewhat well on a 2-platter drive, but imagine an
11-platter drive..
b) Cost. If 2 heads were live at the same time, you'd need 2
amplifiers, 2 DSP's, and a much more powerful input processor in order to
keep track of everything coming in. This would drive up the complexity of
the PCB quite significantly (it's possible, though, to do it with the
technology we have today), adding quite a bit of cost. Of course, people
just aren't willing to pay $2000 for a hard drive in a desktop machine
like they used to.
c) Need. Modern hard drives are really only faster with respect to
sustained transfer rate. This, making up only a small part of the overall
performance of a hard drive under typical loads under typical OS' (random
transfer capability is the single biggest determiner of performance
nowadays), would be the only thing that would benefit from the extra
complexity of doing this. Who's going to pay double the money for a hard
drive that would only perform perhaps 10% better in real life???
To improve the speed of a hard drive in real applications and typical use
would require it to seek out random data more quickly. Again, you're back
to physical actuator movements, and having all of the heads live wouldn't
do a single thing for you.
If you look at the performance of 7200RPM drives, you will find that the
latest high-end Barracudas are only ~20% faster in real-life use than are
the original model ones. This despite the STR on these drives having gone
from ~5mb/second (ST11950N) to ~15mb/second (ST34573LW). It's not a
3-fold increase like lots of people will mislead you into thinking.
Multiple head per platter drives have been tried and didn't do very well
in the marketplace. The Seagate ST12450W was an example of this.
Excellent drive (still a very good performer -- only the Cheetah's are
faster), but a commercial failure for Seagate (most of the R&D came from
their acquisition of CDC/Imprimis's storage unit though). It definitely
was a drive that was way ahead of it's time, having come out sometime in
1994-1995 (forget which year), and still holding it's own even in the
middle of 1998. (how often does 4-year-old hardware do that???)
>Just wondering.......
Richard Krehbiel
>If you look at the performance of different harddisks within the same
>productfamily, eg Maxtor DiamondMax 2880, the 1 platter disk is only
>marginally slower that the 2 or 3 platter harddisk.
>
>simultaneously on multiplatter HDs? That way performance would double on
You can get this performance by using striped sets. There's hardly any
motivation for doing it any other way. Since that would make the drives
cost more for the duplicated data I/O logic, the HD vendor would suffer in
price comparisons.
--
Richard Krehbiel, Kastle Systems, Arlington VA USA
ri...@kastle.com (work) or ri...@mnsinc.com (personal)