At 9:35 Uhr +0100 21.12.1999, Bruce Anderson wrote:
>A IIci should do ok.
>For best performance disable all forms of in kernel compression.
>( My IIci alwase seemed to fall behind if I asked it to do vj )
>ie: add to your options file or /etc/ppp/peers/  file
>novj
>nodeflate
>nobsdcomp

???

I cannot comment about the cost/benefit of VJ IP header compression, but if
your ppp peer is capable of either deflate or bsd compression, you are way
better off when you use that and disable all modem hardware compression
("AT %C0"). Not only is the in-kernel compression more efficient than the
modem's, it also reduces the interrupt rate. The time-critical interrupt
processing is much more expensive per byte than a non-time-critical data
compression/expansion.

That way, I frequently see 4 kBytes/sec when loading text web pages through
a 14k4 modem on a 19k2 serial line.

>Your modem should have a  S register that controls
>RTS-to-CTS delay - (my Zoom 56K uses S26=0).
> set it to zero to reduce overflows.

My modem defaults to "CTS always on".

>Use the modem port rather than the printer port

Not true for NetBSD. MacOS does only poll the modem port when interrupts
are disabled (e.g. in the .Sony floppy driver).

>(no fifo on the printer port).

Huh? The Zilog 85c30 contains two identical ports.

>The real disadvantage is the limited speed of the serial port.
>(57,600bps or 5.625KB/s)

I run an ISDN modem on a IIci with a port speed of 115k2 and data rates of
up to 7200 Bytes/sec (from 8000 max.). As long as you can avoid disk
activity and as long as you don't need any other services (DNS, Appletalk)
from the box during data bursts, you are fine.

>On Mon, Dec 20, 1999 9:00 AM, Jean-Francois Fortier
><mailto:jffortier@cgocable.ca> wrote:
>>My question is:  Will I have exactly the same result if I try it (ppp ddr
>>nat) on macbsd 68030 box.

Again, as long as you can avoid disk acticity, an SE/30 should be able to
keep up with 57k6 serial-to-ethernet just fine.

	hauke



--
"It's never straight up and down"     (DEVO)