[NTLUG:Discuss] OT: SCSI hard drive sizes...

Robert Pearson e2eiod at gmail.com
Mon Oct 23 20:38:31 CDT 2006


On 10/23/06, steve <sjbaker1 at airmail.net> wrote:
> Robert Pearson wrote:
>
> > All SCSI, both parallel and serial (SAS), spin 2x faster than SATA, have
> > much more rugged construction, are dual-ported, have longer MTBF and
> > longer warranties (5 yr vs. 1 yr for SATA).
> > All of which cost money.
>
> When the drive platter spins more rapidly, each magnetic 'bit' has
> less time to induce a voltage in the drive head.  In order to get
> enough signal from the head for the electronics to detect, the bits
> have to be physically larger so they spend more time under the drive
> head.  Larger bits mean less of them.
>
> (Actually, it's probably more due to the duration of the voltage than
> the amount of voltage...but you get the idea).
>
> Hence there is a straight trade between speed and capacity.  You double
> the speed - you halve the capacity.
>
> I presume that some marketting person decided that SCSI buyers are
> more interested in performance than capacity - with the reverse
> being true for PATA/SATA buyers.

What you say has been a factor in reducing capacity.
I'm not sure how perpendicular recording, the new "hot" recording
technology, works exactly. The capacity is increased by stacking the
bits vertically, as I understand it.
I didn't want to get down in the details too much.
I'm pretty out of date on media read/write technology.

I will pass on that the Maxtor 300GB SAS drive has additional platter(s?)
to get its capacity increase. Now if you add perpendicular recording?
The review specifications on performance are:
"This Maxtor Atlas 10KB is a 300GB, workstation class hard disk drive.
Compared to other workstation class hard drives on the market, it is
relatively expensive at around $865. Compared to a typical hard drive,
its maximum internal data transfer rate of 89 MBps is very fast.
Its burst transfer rate (which is how fast the drive moves data already
in its cache into your computer) is 600 MBps.
This is a 10000 rpm drive and has a 16MB buffer.

They do not give sustained transfer rate or any application testing.
The rule still quoted by everyone is,
"If you need Enterprise class performance, SAS is the only way to go".

[Info Note] - Speed Limit of the Information Universe
If you could sustain 89MBps that would be:
1) (89MBps x 3600sec/hour) / 1x10[9] = 320.4GB/hour
2) (600MBps x 3600sec/hour) / 1x10[9] = 2.16TB/hour

For 1) this means the 300GB drive could be read in one hour, backed
up or migrated. It will be slower if any writes have to be done.
For 2) the drive could be read in 30 minutes, backed up or migrated.
Slower if any writes have to be done.

In a "one request at a time" environment this drive will do well.

In a "contention" environment disk transfer rate performance will
depend entirely on cache hits. This will not go as well.

The Speed Limit of the Information Universe
I/O Bandwidth Spectrum looks something like this:

  Performance Oriented (IOPS) |                   | Bandwidth (MBps)
<==========================================>
  OLTP                                                                   Backups
  Search-Find-Obtain (SFO)                                  Multi-media
  Hashing, Indexing                                         Any
streaming operation

Then the I/Os and the impact on CPU bandwidth. Interesting scenario...
Just how fast can Information be delivered to the requester?



More information about the Discuss mailing list