ISA MFM disk controllers
ethan.dicks at gmail.com
Wed Jun 24 14:10:01 CDT 2009
> On Jun 24, 2009, at 10:06 AM, Mr Ian Primus wrote:
> It does work, I just wonder if changing the interleave could make it any faster. I mean, the computer
> isn't particularly fast in any case, but anything to improve hard disk performance is a good thing.
The optimal situation is that you have a controller that can handle
1:1 interleave (i.e., "none" - the sectors are in sequential order
around the track) and a machine that can keep up with that sustained
transfer rate. Where interleave comes into play is if your controller
and/or host is too slow to keep up with one sector after another, so
the sectors are staggered to give the machine time to be ready to
transfer the next block as it's about to pas s under the read head.
The worst case is when your interleave is too "small" for your system,
the next block passes under the head before the system is ready to
handle it, and the system has to wait a full revolution of the disk
for the next pass.
Since someone else mentioned "28ms", it minds me to mention there are
two sets of delays that are now usually grouped together... one is
seek time, which is what 28ms is one measure of, the other is
rotational latency, which is where the interleave factor comes into
play. No matter what your interleave factor, if your drive is rated
at "28ms average seek time", it will take that long to move the head
in and out to the right track. Once the disk is on track, because
it's going to be over a random sector, the average rotational latency
is half the time it takes the disk to spin once - which is why there's
such a push for 10K and 15K drives now (in the old days, 3600rpm was
doing pretty good). _On top of that_, interleave is part of what
determines how long it will take to get to the second then the third
then the fourth sectors on that track.
So if your interleave is badly matched for your machine and disk
controller, you may see some poor sustained transfer rates, but if
it's only a little off, you'll gain back some lost performance, the
amount of which is largely dependent on your disk speed (slow disks
mean better performance recovery, but you still have a slow disk, thus
high rotational latency to begin with).
ISTR a certain obsession with this amongst DOS users back in the day,
but perhaps that was because it was reasonable to expect your files to
be somewhat linear on disk and because hardware performance varied so
widely. Over in the DEC world, you pretty much got what you got, and
with the exception of the RM02/RM03 differences (different rotation
speeds for different systems that could handle different amounts of
disk throughput), we didn't obsess about drive performance minutiae
except when were deciding what disk to buy next.
More information about the cctalk