Hi,

> It would seem, in theory, that a "winmodem" type of device
> could provide better performance since it's data path into
> the system is across a parallel bus and not through the
> serial port.  Also, if the "winmodem" were to DMA it's
> data into a region of memory and then interrupt the CPU,
> it would seem that the system processor could get an
> interrupt per packet instead of an interrupt per
> character...Just a thought....

Don't forget that the data path for serial ports is still
parallel - it's the UART's job to do all the gubbins to get
it like that though.

The IRQ rate of a soft modem would be lower, but the
processing overhead would add latency and increase CPU load.
IMHO, latency is an unacceptable trade-off (being a Quake
3 Arena player :-)

Would you rather handle:-

   a) one byte every 2.27x10^-5 seconds with no processing
      latency and a single, simple driver task with a few
      lines executed (not time critical task).  FIFOs will
      reduce the urgency of handling the reception of data.

   b) one *huge* chunk of data every 1ms with loads of
      processing latency and context switching required
      to handle the driver and dsp task with a few
      thousand lines of code each irq.  This would
      probably have to be complete before the next
      batch of data arrives (time critical task).
      Give a task a high priority and it will lose
      data.

I'd go for A.

- Chris.