I found your graphs very interesting since it made it obvious that I
haven't been paying close attention to changes in hard drive technology
for quite some time. It was clear from the change in performance with
track radius that this drive has something like "constant linear
velocity" (CLV) as is used in CDs and DVDs where the bit density is
constant and the speed of the disk changes to keep the linear velocity
under the heads as constant. Older hard drives use "constant angular
velocity" (CAV). The disk rotation is constant as is the number of bits
per track but bit spacing on the outer tracks is greater. This keeps
the bit rate passing under the heads as a constant but underutilizes the
recording density capabilities of the greater circumference of the outer
tracks.
A little googling told me quickly that newer hard drives hadn't
converted to CLV. They still maintain a constant disk speed but take
advantage of greater track circumference by dividing the tracks into
zones in a technique called "zoned bit recording" (ZBR). Sector counts
per track increase as you move from the inner to the outer zones. Bit
frequency passing under the heads also increases as you move outward but
is constant within the zone.
<http://www.pcguide.com/ref/hdd/geom/tracks_ZBR.htm>
Your analysis of the speed drop-off as due to latency is incorrect since
you assumed that the highest performance was achieved on the inner
tracks. In fact, the highest performance is achieved on the outer
tracks and, unlike CDs and DVDs which begin filling the disk from the
inside out, the hard drive works from the outside in. Latency is only
related to spindle speed and not to recording density. It's constant
anywhere on the disk. However, latency really only relates to the first
seek trying to find the starting sector for some data request. If the
request is large and the data continues on another track vertically up
or down the cylinder the transfer can continue with only a head switch
with no latency involved. Likewise, if the data is on an adjacent track
served by the same head the only delay is track-to-track head
positioning which is on the order of 1ms or less. Again, there is no
latency involved due to cylinder and head skew. Cylinder and head skewt
offsets the starting track on the next cylinder to a position that will
place it under the head just as the head gets there instead of just
having passed by and having to wait for it to come around again.
In any case it behooves us these days to put files which require the
highest performance on the outside of the disk on the lowest cylinder
numbers. There is probably some disk utility somewhere that can monitor
file usage and defrag your disk while moving high frequency usage data
to the outer zones. But I don't know what it is. Maybe it's part of
what you pay for in a server version of the OS.
Chuck Norcutt
Moose wrote:
> C.H.Ling wrote:
>> 5x! Did you ever made a real comparison? I got 22MB per second for USB and I
>> can't get over 30MB with SATA to SATA (two hard disks inside the same PC). I
>> believe the HD sustain rate and other overheads in the computer limited the
>> data rate a lot.
>>
>
> I have not done any careful comparisons, as a big difference was readily
> apparent just watching the progress of a copy operation. I've now done a
> very simple one. I've compared only read speeds for small pieces of data
> - for the simple reason that my free version of HDTune only does that. I
> think the results are instructive.
> <http://www.moosemystic.net/Gallery/tech/Misc/WDC1500_eSATAvsUSB2.jpg>
>
> Unfortunately, the program also rescales the graph height, so you need
> to look at the vertical scale and/or the numbers on the right. Both
> tests are of the same drive in the same docking station. The top one is
> eSATA. It shows the typical shape, flat at first, then tapering down as
> latency becomes greater as the circumference of the tracks increases
> farther out on the platters. In fact, it's a better than average plot,
> with a long high speed section. As it does slope down slightly right
> from the start, I believe the limitation is the disk itself, not the
> connections.
>
> The second plot is of the same equipment, but using the USB 2.0 HSE
> connection. The dead flat plot is a clear indication that the USB
> connection is the limiting factor in speed. So at least for read speed,
> the 5x raw spec is optimistic for this drive. On the other hand,
> 85.4/27.2 is still 3x. To date, I've only filled part of the 1.5Tb
> drive, so I am still seeing something like the 90 MB/sec part of the curve.
>
> But again, the limitation is the drive, not the eSATA connection, so
> this test doesn't say what that might be capable of.
>
> C.H.Ling wrote:
>> So when we are talking about external drives for backup, then writing time
>> is also very important, eSATA does not has that much advantage here.
>
> I assume you meant to say "writing time is also *not* very important".
>
> Sure, but the price difference between USB only and USB & eSATA was only
> $US5. So why not go with the faster interface? The docking station can
> handle any 3.5" or 2.5" SATA drive. The possibilities for other tasks,
> where speed could make a difference, are there.
>
> C.H.Ling wrote:
>> My experience is if you have a large set of files to copy the result is
>> quite accurate (did many tests in the pass twenty years), I bet no more than
>> 20% variation in most cases unless your dirve is almost full.
>
> I've never measured that. Frankly, I doubt is I will. It's obvious to me
> from experience that eSATA is much faster. Just how much I don't need to
> quantify.
>
> As Chuck said, such a test is only valid if all drives are defragmented
> before testing.
>
>> To gain more speed you better go with a 10000rpm drive or a RAID setup.
>
> That is much more expensive than a simple eSATA appliance. And the
> higher speed drives don't come in the largest sizes, at least that I've
> seen.
>
> Another other factor to consider is that the 1TB and larger drives use
> vertical recording, so the bit density is higher and any given read
> speed can generate a higher output speed. The drive I tested has that
> advantage. So I added a test of a WD 500GB drive without vertical
> recording to the example. This is an internal drive, directly connected
> to the MB SATA header.
>
> As you can see, the graph starts much lower and drops much faster than
> the 1TB drive. So it appears that the advantage of eSATA may be greater
> for 1TB and larger drives.
>
> Moose
--
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/
|
