I just bought a Antec MX-1 hard disk enclosure to house a WD50000AAKS. When I rebooted my PC, I saw that the Marvel driver was installed during boot time.
I copied 21GB from my local drive to this external drive and it took 6 minutes. Not even close to the 3 GB/S as per SATA-2 standard. When I connected the HDD to a USB-2 port, the same data would take 14 minutes to copy.
The ESATA transfer rate does not even come close to the SATA standard of 3 GB/S. Is this a problem with the Marvel driver? If I use a ESATA bracket to connerct the external HDD to one of the 6 internal SATA ports, would this improve the transfer rate?
Umm you realize that you will never max out the SATA bus speed. Hell todays drives couldn't max out IDE. Not only that but SATA does not do 3 G Bytes (capital B denotes bytes not bits) is 300MB/s for SATA II. Now with that hard disks are limited by the fact that they are a mechanical system, their rotational speeds can not approach the speed needed to saturate the SATA bus (the typical rotational speed for desktop drives is 7200RPM and WD Raptors are 10K). Most new drives don't reach beyond 75-80MB/s. If you want faster you have to go SCSI with rotational speeds of 15K RPM, or do RAID 0. The SATA bus speeds are only worth considering when writing small files (small enough not to overflow the buffer) or reading small files that happen to be in the drives buffer. Any sustained write/reads will have to access data from the drive platters themselves which is limited by it's access time + rotational speed.
So to transfer your data (21GB) at a sustained transfer rate of 65MB/s (typical), it would take approximately 5 minutes 40 seconds. Given the fact that I don't know where on the platter you data lies (outside of the platter gives fastest speeds) and I don't have knowledge of how fast your HDD actually is, this is an educated guess. Saying that, it seems consistent with what you have stated. As for USB, the very maximum you can transfer over USB 2.0 is 60MB/s. That is ideal based on the maximum bandwidth of 480Mb/s, you will never reach that speed. It's more than likely that you might see 30MB/s, so USB will be much slower than ESATA (take approximately twice as long).
You are right. I forgot about the bit/byte differences.
I use HD Tune 2.54 to benchmark my 2 drives:
HD Tune: WDC WD7500AAKS-00RBA Benchmark (SATA)
Transfer Rate Minimum : 48.4 MB/sec
Transfer Rate Maximum : 97.1 MB/sec
Transfer Rate Average : 76.8 MB/sec
Access Time : 13.8 ms
Burst Rate : 124.4 MB/sec
CPU Usage : 1.1%
HD Tune: WDC WD5000AAKS-22TMA Benchmark (ESATA)
Transfer Rate Minimum : 39.0 MB/sec
Transfer Rate Maximum : 81.9 MB/sec
Transfer Rate Average : 66.4 MB/sec
Access Time : 13.4 ms
Burst Rate : 99.9 MB/sec
CPU Usage : 1.6%
The internal SATA port (South Bridge as opp0sed to Marvell) is faster. Maybe I should use a ESATA bracket to connect the external drive to one of these SATA ports.
... or, if this is your "source drive" --
HD Tune: WDC WD7500AAKS-00RBA Benchmark (SATA)
-- please confirm.
Writing is predictably slower than reading, and
when copying a large volume of data, like 21GB,
there will be an accumulation of access times
on the source drive that will result in lowering
the raw data rates overall.
If the target drive is empty, access times
to position the read/write heads will probably
not accumulate to much of a factor, because
the heads will move from track-to-track;
nevertheless, there is still the issue of
rotational latency, which is always one-half
of the rotational speed.
After the read/write heads are positioned
over the desired track, the platter must
spin an average of one-half rotation
before the heads can begin writing,
or reading (as the case may be).
On a 7,200 rpm HDD, one-half rotation
requires (60 seconds/7200 rpm)/2
or one-half the time for one revolution.
> If I use a ESATA bracket to connect the external HDD to one of the 6 internal SATA ports, would this improve the transfer rate?
Give it a try, and let us know, please.
Be sure that your eSATA cable is a quality cable,
because eSATA generally requires a higher line voltage,
to satisfy the longer cable lengths of the eSATA standard.
The Marvell controller should be using the proper line voltage,
but the main SATA ports may NOT if they were NOT built
to meet the eSATA standards.
Here's another source of overhead that
users generally overlook: NCQ
(Native Command Queuing).
An eSATA port with Marvell controller
may default to enabling NCQ.
However, for purely sequential writing
to an otherwise empty HDD, NCQ will
add computational overhead that is
not strictly needed.
In other words, the controller logic
will compute the optimal next track
when the read/write heads are
already at or near that optimal track!
Thus, that extra computation is wasted.
Several users have reported SLOWER
I/O times on their storage subsystems
with NCQ enabled (believe it or not!)
It's only when files are much more fragmented,
and the controller has numerous I/O requests
pending in its "queue," that NCQ will have a
real opportunity to optimize the sequence of
I/O requests to process.
Thanks: I inferred as much, but only after I wrote my first reply.
I still maintain that one important reason for the overhead
was the cumulative access time at the source (SATA) drive
i.e. head seeks and rotational latency.
Here, with our database, we found that keeping a 2GB+ database
updated across several backup copies, each storing 75,000+ discrete
files, was easiest by using XCOPY in Command Prompt, as follows:
This only updates component files that have changed,
and it works across a LAN quite nicely too.
A command file can use a command line option, like this:
xcopy supremelaw.org e:\supremelaw.org /s/e/v/d%1
... then, the "/l" option can be appended, which reports
the changes that XCOPY would make, but it does not
actually update any files. This option is good to run first,
if you're not sure which files have changed.
If you plan to be copying large volumes of data
on a regular basis, a RAID 0 is highly recommended,
particularly using modern on-board RAID controllers
like Intel's ICH9R I/O Controller Hub.
For example, on the ASUS P5K-E/WiFi-AP motherboard,
there are two (2) on-board eSATA ports at the rear panel
which can be configured with several RAID options.
On our dual-core development machine, we started
out with 2 x 500GB Western Digital "RAID Edition 2" HDDs
configured in a RAID 0 array, and it really moves data
much faster than a single HDD ever could:
This is a nice combination: 1TB plus moderately fast speed
for less than the cost of a single 1TB HDD, even with PMR
and the very latest technology.
Their interface still runs at 150MB/second;
their cache is only 16MB instead of 32MB; and,
they do not use Perpendicular Magnetic Recording ("PMR" )
(not yet, anyway).
A Raptor "update" is long overdue, imho.
The one measure where Raptors still excel
is in access time, and that's no surprise,
given a known rotational speed of 10,000 rpm,
and platters with much smaller diameter.
The smaller diameter has a big cumulative
effect on the average distance the read/write
heads must travel for successive I/O requests.
I'm told that Western Digital chose smaller
diameter platters also because the larger
platters would have necessitated larger
main bearings, due to the added centrifugal
force of those larger platters spinning at
10,000 rpm instead of 7,2000 rpm.
Nevertheless, many 7,200 rpm SATA/3G HDDs
have caught up with the Raptors,
as far as raw transfer speeds are concerned,
measured in average Megabytes per second.
Some users think that the Raptor makes an
excellent host for XP's C: system partition,
because program launch times tend to be
shorter.
I agree, but I also think that launch times are
not the best place to focus optimization efforts.
Compare shaving a few seconds off a
program launch time, with optimizing ALL I/O
to a browser cache that occurs during a
very long IE7 session lasting ~4-5 HOURS!
Instead of speeding up program launch times,
we've had much more success, overall, by
moving frequently used files into a RAMDISK
hosted by the RamDisk Plus software from
www.superspeed.com .
Laptops are helped in particular by this setup
if they can be easily upgraded from 1GB to 2GB
of RAM: moving the IE7 browser cache into
a 512MB RAMDISK also results in off-loading the I/O
load required of relatively slow 5,400 rpm laptop drives,
reducing wear and prolonging their useful life.
If you want the fastest possible storage subsystem,
15,000 rpm SAS drives are rated at ~125MB/second.
Moreover, 4 can fit into a single QuadraPack Q14,
requiring only one 5.25" drive bay, and configured
as a RAID 0 on a modern controller, for maximum
throughput.
