Joe wrote:
>So, can anybody think of a *good* reason why the FIFO receive
>interrupt threshold should be set to 8??? I know this theoretically
>reduces the comm's overhead a bit, but in practice, this seems to
>be the cause of poor comm's performance (I know I am not the
>only one to have this trouble!).
Perhaps this is very good in Linux? (epppd is a mixture of programs
originally in Linux - parts even made by Caldera)
>I understand than Bernie is rewriting "epppd(d)" to reduce it's memory
>needs
yes.
>(just out of curiosity, how is this going to be achieved?).
Two parts:
1. Reorganize so that code that's useless (ex. functions calling functions
are turned into macros) isn't used anymore, make small diffrence (1-2K IIRC).
2. Remove the parts that Arachne never uses (command-line, most of the
commands in pppdrc.cfg) (8-9K currently, but I'm not finished yet - work
was stopped since I didn't have the latest source code for Arachne earlier)
I haven't worked on it for over a month now but it's working perfectly for
me (I use it almost daily) and uses 10K less conventional memory. Now to
make it more interesting I'll try and look into how cloaking.exe is
working, if I can get it to load in XMS instead of conventional (well a
small stub will be left, 1-2K or so) it will be much better. Also I'm
perhaps going to make CCP work (now the communication channel is slower
than for instance Win9x if you specify compression in the later), but that
will generate more memory used so I better get cloaking working first.
>So,
>I would like to propose that this change to the FIFO receive interrupt
>threshold also be implemented. Any thoughts on this ... ?
Would that really increase speed for most people? Perhaps a second version
could be made instead?
