BSD 4.3 performance optimization and the DZ-11

[Jeff Siegal]

Can someone explain the performance optimizations that are included in
the BSD 4.3 DZ-11 driver?  I know there is some sort of automatic
switching between interrupt-driven input and silo-read polling.  How
(and how well) does it work?  Are there any other optimizations?  Any
to deal with output, for example? 

For high-volume input, would one expect better performance (CPU
loading per character, not data rate) using a DZ-11, or a telnet
connection?  How about a straight TCP connection? 

Jeff Siegal