ngc@ff.iij4u.or.jp wrote:
> - driver for Dayna PORT Ether/Link
>   http://www.ff.iij4u.or.jp/~ngc/data/if_dse.c.gz

hi hiroshi,

two things i've noticed about the design of your driver:

the transfer speed is limited to around 80 kbytes/s due to the delay of 
at least one tick (1/60 s) between each poll for a 1500 byte package. 
the source says "a zero delay actually worsens performance" but this 
sounds a little weird. i'll try it and get back with the result.

the transfer speed is further limited by the "nominal" polling speed, 
since i believe the driver starts polling at that rate immediately after 
reading the maximum number of packages. seems like the best guess after 
having read 10 packages would be that there will be more available, so 
the driver should poll at the "minimum" rate at least once just to 
check? i noticed this when i changed the "nominal" speed from 80 to 500 
ms, the same as the macos driver, since this helped with my "random bus 
errors at high load" problem. this caused transfers to run at only 14 
kbytes/s, and you could actually see ten hashes, pause, ten hashes, 
pause, in ftp. i'll try to fix this too, but i thought it would be nice 
for anybody else working on this driver to consider...

that's it for now. thanks for your time. which list, if any, should this 
discussion continue on? i've cc'd port-mac68k and tech-kern, but i have 
no idea whether either of them are right...

	/matt