On Fri, 17 Mar 2000, Iljitsch van Beijnum wrote:
> On Fri, 17 Mar 2000, Fred Wright wrote:
>
> > Well, 115.2K teeters on the edge of usability with the built-in serial
> > port, so it doesn't take much to push it over. :-) Of course if the driver
> > is doing something obscene like copying all the data under a Disable() it
> > should be shot.
>
> I forgot to mention that the actual connection in this case was at 57600
> bps. The 1200 has no trouble routing between the serial port and the
> ethernet at this speed. Yet the 1200 has a 50 MHz 68060 and can't do
> 115200, while the 3000 with the original 25 MHz 68030 can (at least when
> there's no ethernet).
CPU power isn't the main issue. Contention for custom chip sccesses is.
People often forget that accessing chip *registers* is just as expensive
as accessing CHIP *RAM*. Every interrupt needs to do it, every Disable()
or Enable() needs to do it, etc. With a maxed-out native display, a
single custom-chip access can take over 50 microseconds. Compare this to
the requirement to service the serial port every 87 microseconds at
115.2K to avoid character loss.
Fred Wright
--
To unsubscribe send "unsubscribe miami-talk-ml" to
"[EMAIL PROTECTED]". For help on list commands send "help" to
"[EMAIL PROTECTED]".
