dean said:
The specs are are 2.54GHz P4 and the pics are on a WD 7500AAK external
USB 750GB drive. I've compared it to the 76GB internal drive (sorry
not sure what brand) and there's no difference in speed.
If I reverse the images and load them all from memory cache instead,
its just less than twice as fast (not quite 2x faster), so it seems I
am indeed (as mentioned) slowed by my system - I bet that is all I can
possibly get given my processor. Maybe a new PC is in order.
Am I wrong or are there no Ultra320 SCSI cards out there for PCI (not
PCI-X or PCI-E) slots?
Taking a look around, I cannot quickly find a PCI card, with 32 bit bus
connection, offering a U320 bus. There are plenty of cards offering
U160, which is closer to the 120MB/sec limitation that comes with a
PCI 32 bit, 33MHz bus. Chances are, the U320 drive you are buying, could
work at a number of different SCSI bus transfer rates, all the way down to
20MB/sec or maybe even lower. So the declared drive speed, does not mean
you need to buy a U320 card. The interface on the drive will probably be
LVDS (low voltage differential swing), and that level and type of interface
has existed for years. A U160 would still support the sustained media
transfer rate of the fastest disks. The main benefit of the U320
transfer rate, when using a single disk, would be the ability to
"burst to cache" on the SCSI disk's controller.
There are cards which use one of the faster PCI standards. For example,
you can have 32 bit PCI cards operating at 66MHz. But you are unlikely
to find support for that on any old desktop motherboard. Another option,
would be cards that are 64 bits wide. That means a longer connector on
the PCI card.
This is an example of a longer card.
http://discountechnology.com/Adaptec-2120S-U320-LVD-SCSI-RAID-Controller?sc=2&category=31#photos
Once you get into this league, there are some other things to consider.
The PCI-X higher speed cards, for example, operate at 3.3V only. Some
people get a rotten surprise, when they think they found a bargain
somewhere, and then they notice that the voltage slots cut in the card,
prevent it from fitting in their motherboard.
In this closeup picture, you can see three slots cut. The right most
one, separates the two 32 bit parts of the interface. If you plug a 64
bit card, into a 32 bit slot, it can still work, and 32 of the contacts
are left "hanging in the air". The rightmost slot is there to leave room
for the end of the connector. This is a U320 card, but if plugged into a
32 bit, 322MHz PCI desktop slot, the max transfer rate achieved will be
120MB/sec or a little less.
http://files.discountechnology.com/products/ADP-2120S-BN-FW/ADP-2120S-BN-FW-image1.jpg
Next to the 32 bit separator slot, is the 5V key. A lot of desktop
motherboards, would have that key in place. It means the VIO selected
for the motherboard is 5 volts. A motherboard designer has to make some choice
of either 5V or 3.3V when the motherboard is designed, and 5V is compatible
with a lot of older stuff.
The left-most slot cut in the sample card, is for 3.3V. Since that
card has both 3.3V and 5V VIO capability (i.e. universal), it can
work with motherboards that choose to supply 3.3V or 5V. So that
card should be able to plug into a desktop, and also into the faster
slots (like PCI-X or PCI 64) on a server motherboard.
You have to be careful with this stuff. At least Adaptec, provides information
for some of their products, so you know whether the card will work for sure
in a PCI 32 bit slot or not.
So if you buy a U160 card, with an ordinary PCI slot connector on it,
then there should be no problem plugging it into a desktop motherboard.
For some of the other types, you can either rely on the slots cut in
the board, as an indicator, or check the manufacturer's web site, for
confirmation of interoperability. Notice, our example ADP-2120S card
above, is listed as "64-bit or 32-bit *" and "66 or 33 MHz" in the following
web page. The 33MHz part is reassuring, since that is the speed of a
desktop PCI slot. The card has no voltage limitation shown, and we knew
that since it had both the 5V and 3.3V slots cut in it, for VIO choices.
The asterisk seems to suggest they are looking for a more recent PCI slot
standard (2.1 or 2.2) and PCI 2.1 has been around for some time.
http://adaptec-tic.adaptec.com/cgi-bin/adaptec_tic.cfg/php/enduser/std_adp.php?p_faqid=1903
So you can grab a card that has a 64 bit connector on it, but it would
pay to do some research first, to confirm it would work. There is the
odd case, where the plastic frame on the 32 bit connector, is too wide
to fit into the separator slot in the 64 bit card, but I don't seem to
see that mentioned any more.
One test I've used before, to eliminate disk speed from an experiment,
is to use RAM disk software. That makes storage space, from a section of
RAM. That is only feasible, if you have lots of RAM to play with. For
example, I wanted to see how fast a network connection I could make
between two computers. I had a Gigabit connection, and both computers
had Win2K. Each computer had 1GB of RAM, so I made a 100MB RAM disk on
each computer. I was only able to achieve 40MB/sec, only 1/3rd of the
available gigabit link speed (which is what that particular OS is known
to be able to achieve) with the setup. By using RAM disks, I was attempting
to eliminate the storage devices from messing with my experiment. The
RAM disks I've used, were not the cleanest things to install and remove,
so they tend to be "creaky" pieces of software. Microsoft has a sample
RAM disk implementation, and some of the available RAM disks are based
on the sample implementation. I'm not sure if anyone makes a "smooth"
and "flexible" RAM disk implementation, that is free from flaws.
If you are processor limited, then the solution to that should be
more straight forward.
Paul
