The original posting mentioned DOS using mode3,
but the numbers shown (240us/cycle) were slower than mode3 (180ns).
> My magic recipe is:
>
> hdparm -u1 -m16 -a16 -A1 -c1 /dev/hda
>
> Though your mileage may vary, of course.
...
> Notice how %CPU columns have lower values, even with better I/O
> throughput. Tested disk is IBM DAQA33240 (3.2GB).
The %CPU is not a meaningful measure here, though it probably
does reflect lower overall CPU usage in this example (think about it..
with an infinitely fast disk (eg. ramdrive), CPU will be 100% simply
because I/O takes no time..)
In this case, the reduced CPU is attributable to the -m16 parameter,
which basically cuts the IDE interrupt overhead down to as little
as (in this case) 1/16th that of the default setup.
-- mlord@pobox.com The Linux IDE guy