> 
> On Mon, 3 Nov 1997, Bill Studenmund wrote:
> > No clue. What happens if you stick in a jra _ASM_LABEL(rei)?
> 
> Well, ping times are much better:
[snip]
> (see http://www.bga.com/~khym/netbsd/ttcp.html for numbers with the rte
> instead of jra _ASM_LABEL(rei))... The quick summary is 869.69KB/s
> receiving and 860.80KB/s sending.
> 
> Serial performance is perhaps slightly worse than before, but not
> significantly, in my opinion. I ftp'd a file over a 57.6kbps PPP
> connection (null modem, so it really is 57.6k) while flood pinging the
> Mac, and got occasional ring overruns, but I was getting those before too.
> I don't know what the change would do on slower machines like the Mac II
> though.

Two things:

If you're getting ring overruns, try running w/ cdtrcts. It will ask
the other side to stop sending when the ring gets near full. Make sure
HSKo (DTR) gets wired to RTS on the other computer. If it's a mac,
then a mac printer cable will work just fine.

Also, I think the serial interrupt doesn't mess w/ the softint checking
as we get one interrupt for almost every character. Thus we might spend
a lot of time in the interrupt service routines. Since one call to the
software completion routine will read in a lot of characters, we gain
efficiency by waiting.

As we just got dynamic interrupt level support, the best thing to do might
be to leave the serial stuff alone, and put ethernet and serial I/O on
different interrupts.

Take care,

Bill