On Wed, 8 Jul 1998, Hauke Fath wrote:

> At 4:49 Uhr +0200 08.07.1998, Michael G. Schabert wrote:
> >>On Jul 4, 12:37pm, Roger Fischer wrote:
> >>>On old Macs, (centris) I seem to remember that the serial port speed
> >>>should not be set above 38400 because it just can't keep up.

My IIsi does fine at 38400. I'd expect a centris (040) would be fine at
57600.

> >Similarly, to my knowlege, NetBSD will have a harder time keeping up with
> >fast serial ports activity, because of the way the Mac's interrupts are set
> >up (a la clock losing time).
> 
> No. The SCC interrupts at level 4 which is the highest level available on
> most Macs (besides the NMI switch). Actually, for the three byte fifo and
> the cpu power available, serial performance on NetBSD/mac68k rocks - thanks
> to Dr. Bill. I run my IIci at 57k6 (ppp and uucp) with nothing more than a
> silo overrun every few minutes.

The real problem is not getting bytes out of the chip, but getting them
processed (resulting in floods).

If you're getting silo overruns or floods, you might need to either turn
off ppp compression, or lower the baud rate. one silo overrun every few
minutes is ok. But each error means an MRU-sized packet must be resent.

Just so everyone understands this, a one byte in 1000 error means about
one out of every two packets (of 576 I think by default) dies. With
re-transmission, that effectivly means each packet gets sent twice, so
you're seeing half the link efficiency - you might as well lower the baud
rate.

Also, it's better to run w/o ppp packet compression, I've found. Certainly
so if your modem's doing compression too. The ppp packet
compression/uncompression tends to constapate character processing and can
back things up. Header compression's good, though.

Take care,

Bill