Want to protect your cyber security and still get fast solutions? Ask a secure question today.Go Premium

x
  • Status: Solved
  • Priority: Medium
  • Security: Public
  • Views: 764
  • Last Modified:

# of Hard Drive platters vs. average seek time

I was review technical specifications for a Hitachi Disk drive E7K1000.  First, I assume that 1D, 2D & 3D refer to the number of disks/platters that are in the physical drive.  Is this correct?  Assuming that is correct, then as the number of platters increases in the drive the average seek time goes down.  This seems opposite to conventional thinking.  Normally as you add additional heads (weight) this increases the inertia of the head assembly and it takes longer to move the head from one place to another.  If you have ever had to work with any positioning system, you always want the positioner to be as light as possible to have the fastest switching time.

Can someone explain why they show a decreased seek time with more platters?


(http://www.hitachigst.com/tech/techlib.nsf/products/Deskstar_7K1000.B)

Link to specification PDF: http://www.hitachigst.com/tech/techlib.nsf/techdocs/CF6C5D47F5BCE65B862574C1007E985D/$file/DS_CS_7K1000.B_Spec_rev3.01.pdf

Thanks!
AverageSeekTime.jpg
0
TedgCl
Asked:
TedgCl
  • 5
  • 3
  • 3
  • +1
1 Solution
 
Andrej PirmanCommented:
Just to comment on your logic:
your assumption of head movements beeing a bit slower when having more heads is kinda correct, but on the other hand, 2 heads read twice as much data as 1 head, so being head for example, 20% slower, it has access to 100% more data, which is actual increase.
Also, more heads cover larger area, which is also increase in performace.
And least but not last, speed of read bytes per 1 disk revolution actuall doubles with each head, which is pure mathematical increase in data seek speed.
0
 
Andrej PirmanCommented:
Ups...I pushed SEND too fast (only one head here, hehehe)

What I wanted to explain is pure logic on disks being faster and having lower seek rate with multiple disks inside:

Imagine 1 head reding 1 disk surface.
It reads 1 data bit.
Then it should read next data bit, which is somewhere else on disk surface, not exactly in previous bit neghbourhood.
Disk plate needs to rotate in worst case almost 1 whole cycle to reach the requested data bit.

But if there are more disks and more heads, chances are much much better that next requested data bit is somewhere closer to some other heads, so data will be retrieved faster.
Ok, if next requested data bit is at the same disk plate and the whole disk turn around, then seek rate would be the same.
But all I say is that CHANCES to be faster are only in situation where you have more disk plates.
0
 
akahanCommented:
1D, 2D and 3D don't refer to the number of platters at all; it's a reference to the size/format of the data buffer.
0
 [eBook] Windows Nano Server

Download this FREE eBook and learn all you need to get started with Windows Nano Server, including deployment options, remote management
and troubleshooting tips and tricks

 
akahanCommented:
Actually, hmmm.  I'm not so sure about that now...  
0
 
Andrej PirmanCommented:
Hmmm...as I understand:
1D stays for (imaginary, theoretical) data distribution on disk in a single circular row. It is like reading data from only one row in one single circle.
2D stays for data distributed in MANY circular rows, which is more close to reality, but is valid only for disks with 1 single head and 1 side of disk plate.
3D stays for many 2D arrays, distributed among many disk plate sides or many disk plates, which is most common in reality.
0
 
Andrej PirmanCommented:
Ups...found some readings:
http://www.maths-in-industry.org/miis/167/1/hard_disk.pdf

1/2/3D stands for kinda "Dimensional", and si used in terms of data seek algorithms optimisation. If only 1-dimansioanl (1D) data is calculated within optimisation, results are poor. Better cache optimisation is made by optimised algorithms having calculations for all 3-Dimensional data distribution. Kinda :)
0
 
akahanCommented:
Labsy,

I don't think that makes sense.  See Section 4.5.3 of the PDF posted by the original poster; why should "power on to ready" time differ depending on the layout of the data ?   Similarly, in 6.5.1, why would the weight of  the drive in grams differ?
0
 
TedgClAuthor Commented:
Thank you both for your comments but I don't think we have completely answered the question yet.  

Agree with Akahan regarding the possible definition of 1D, 2D & 3D.  The drive weights and mode transition times seem to suggest that the D's refer to th actual platters.  The more weight you have on your platter assembly the longer it will take to spin up and spin down.  A 1D model drive spins up from 4,500 to 7,200 in 3 seconds where a 3D model spins up in 6 seconds.   This makes sense as the energy input is likely similar for all three models but there will be an increased mass for the 3D drive.  The basic equation for inertial storage is I=k*M*R^2  (K and R are constant for the system but M changes).  Similar you could also model the total kinetic energy which only would differ with the mass component.

It is the same physics that make the chart for the seek times not make any sense.  It may be true that with multiple heads you can be simultaneously reading from multiple platters but we are not talking about data throughput, we are talking about the seek time, the time it takes for the head to get from one random place to another.

I don't agree with the statement "But if there are more disks and more heads, chances are much much better that next requested data bit is somewhere closer to some other heads".  Every hard drive I have ever taken apart has the head assembly as one piece.  This means that the heads don't move independently.  Since the heads are not independent and the seek time is measured by a truly random event, the data could be anywhere on any platter and the heads all as one unit would need to move to the track.  Additionally, there is a table 4.5.2.2 that has full stoke seek.

I actually think most of the seek test are done by asking the drive to move the head positioner to a certain track and then they measure how fast it gets to that position.  

Thoughts?
0
 
TedgClAuthor Commented:
I found this where they talked about 1,2,3 Disk SATA models
Disks.jpg
0
 
Andrej PirmanCommented:
Re:
"I actually think most of the seek test are done by asking the drive to move the head positioner to a certain track and then they measure how fast it gets to that position."

Yes, that's true.
So, if we look stright into physics, it is certain that multiple heads wil definitelly be slower on data seek than single head, due to inertion factor - it has to start and stop higher mass.
But still...as I read those papers, 1D, 2D and 3D again seems to me like abbrevations of 1-dimensional data layout (linear), 2-dimensional (1 disc's surface) and 3-dimensional (data accross multi discs array). Seek algorithms are optimised for those layouts.
0
 
nobusCommented:
and why is it in quiet mode the same then ? strange issue
0
 
TedgClAuthor Commented:
This question didn't get answered.  What I did discover from Hitachi is the 1D, 2D, 3D stands for the number of internal platters.  Couldn't get an answer about 1D vs. 3D ... sorry.
0

Featured Post

VIDEO: THE CONCERTO CLOUD FOR HEALTHCARE

Modern healthcare requires a modern cloud. View this brief video to understand how the Concerto Cloud for Healthcare can help your organization.

  • 5
  • 3
  • 3
  • +1
Tackle projects and never again get stuck behind a technical roadblock.
Join Now