On Wed, 15 Dec 2004 22:27:35 -0500, Paul A. Pennington
<paulpenn_at_knology.net> wrote:
> What kind of interleave performance could you get with an 8-bit SCSI
> controller?
You don't format SCSI drives for different interleaves, unless you
happen to be fiddling with, say, an Adaptec bridge controller with an
MFM or RLL drive on the other side.
What you _do_ get, is the ability to use modernish drives (subject to
capacity limitations of your controller BIOS, esp. in the case of an
AHA1542A) that may have hundreds of K or a few megabytes of on-disk
buffering. With those drives, your application asks for a block, and
the drive itself reads-ahead so the next few blocks are delivered from
the on-disk RAM cache... relieving you from the need to worry about
interleave.
Do you have a good grasp of why older machines had to worry about
interleave in the first place? The controller read and converted
analog signals in real-time, no track buffering. Controllers that are
not analog in nature allow the drive to do whatever it likes, since
the controller is just going to ask for a block whenever the drive
gets around to it. This is not to say that the drive doesn't have an
interleave (older ones might), but it does take the need to interact
with the drive surface away from your list of concerns. The reason
for different values (1:1, 5:1...) have to do with how fast the disk
rotates and how long it takes for the host to be ready to slurp up the
next sector. If your controller or host machine isn't fast enough,
you might as well use a huge interleave to save a full rotation
waiting for that block to come around again. Obviously, the ideal
situation is 1:1 interleave - one rotation means you can slurp up an
entire track without other delays. I don't recall this being a common
situation until the days of the AT. I remember 1:3 to 1:5 interleaves
being ordinary in the hey-day of the 8-bit PC and XT.
So... back to what the orginal poster said, if you can find a
DMA-capable 8-bit SCSI controller and a modernish drive (4GB and newer
would have much larger disk caches than you have physical memory), you
might be able to severerly reduce rotational delay (and seek time as a
consequence of using newer drives than, say, an ST225 w/65 ms average
seeks) enough to make a difference.
-ethan
Received on Wed Dec 15 2004 - 22:31:45 GMT