At 16:03 05/07/03, Mike Whalen - The Computer Valet wrote:
>Someone recently asked me to benchmark their SCSI HDD. They believe
>their HDD access is not up to par.
>
>It's in an NT4 system. The mainboard is a MSI MS-6309 which has a
>133MHz bus.
>
>Inside is an Adaptec 29160N Ultra160 SCSI Controller. The drive itself
>is a Seagate ST318437LW, A.K.A. a Barracuda 36ES.
>
>I've read the drive's specifications (PDF) and it looks to me that the
>drive can push out:
>
>28.1 Minimum
>42.0 Average
>50.8 Maximum
>
>(MB/sec. This is in section 4.2.3. of the PDF. The PDF is available
>here at http://tinyurl.com/b88s
>
>According to SiSoft Sandra's index of C: (which is one of the
>partitions on the drive), this drive is actually pushing between 19.0
>and 23.0 MB/sec.


These numbers are the "data buffer transfer rate to/from disc media
(one 512-byte sector)". That is, the speed of transferring data from
the internal platters to the buffer (solid state memory) inside the
drive itself. A file stored on your hard drive will likely not be
in one single sector. It would be spread across many sectors (maybe
thousands or even millions of sectors) and these sectors are probably
not right next to each other on the drive surfaces.

So these numbers are not the transfer rates of data moving from the
drive to your computer. They are the absolute best case for
transferring 512 bytes. Remember that each time the drive has to
go from one sector to another, it takes maybe 10 milliseconds (the
access rate) and that slows things down. Even if all the sectors are
contiguous, things will be significantly slower than the theoretical
values.

As far as the SiSoft results, I wouldn't trust these either because
of the disk caching that's automatically done by Windows with your
free RAM. The only way to get around this caching would be for the
benchmark software to transfer many files, each file being several
times the size of your physical RAM, and analyze the results with
appropriate algorithms. I doubt that SciSoft does that. The only
program that I've found that does things properly is Adaptec's
Threadmark program, and that's no longer posted on Adaptec's website.
I found a copy of Threadmark 2 at a foreign website. I went to
http://www.alltheweb.com/ and searched for the file name
"thrdmk20.exe". I got one useful hit:
http://www.mv.helsinki.fi/public/HARDWARE/SCSI/ADAPTEC/thrdmk20.exe
I would suggest that you download it now and save it for all time.

SciSoft may be good for comparing the speed of different drives on
one particular system, but not for comparing a drive on one system
to even the same drive on another (different) system.

By the way, if SciSoft is correct, 19 to 23 MB/sec is very good.
That's almost as fast as I got for my 10,000 RPM Seagate Cheetahs.
(These drives have faster transfer rates due to the higher rotation
rate and also have access times about half that of the Barracuda's.)
But as I said, I don't think is is good enough to compare results
from one computer to another. (That's what Threadmark claims it
tries to do.)

Regards,
Bill

            Do you want to signoff PCBUILD or just change to
                    Digest mode - visit our web site:
                   http://freepctech.com/pcbuild.shtml