server.lisp

[<don...@is...>]

 > (transmission speed approx. 300 bytes / sec!!!).

This is much slower than I normally see.

 > the problem seems to be the read-byte/char and write-byte/char loops, like
 > in server.lisp:
 > 
 > (loop for i from start below end
 > 	for char-number below *cycle-io-limit*
 > 	while (output-possible stream) do
 > 		(write-char (char output i) stream)
 > 	finally
 > 		(force-output stream)
 > 		(setf (third out) i)
 > 		(if (>= i end) (pop (output connection))
 > 		   (return-from process-output nil)))

You are correct that this is the bottleneck.  At the end of
server.lisp are some timings.  On a 200MHz machine the cost is
something like 25usec/character, half for the output-possible and
half for the output.  That comes out to about 40K chars/sec.
On a modern machine you ought to get about 400K.  I don't think
this is the reason for 300B/s unless you have a REALLY slow machine. 
Or maybe you're running the code interpreted?

[Sam writes]
 > use
 > write-char-sequence (http://clisp.cons.org/impnotes.html#wr-ch-seq)
 > write-byte-sequence (http://clisp.cons.org/impnotes.html#wr-by-seq)
 > read-char-sequence (http://clisp.cons.org/impnotes.html#rd-ch-seq)
 > read-byte-sequence (http://clisp.cons.org/impnotes.html#rd-by-seq)
 > instead of write-char, write-byte, read-char, read-byte.
 > 
 > or create buffered socket streams
 > (http://clisp.cons.org/impnotes.html#buffered)
 > 
 > note the new :NO-HANG arguments to read-byte-sequence (2.31) and
 > write-byte-sequence (CVS head).

The no hang argument for write-byte-sequence was added specifically
for this reason.  I have a new version of server.lisp that uses it
and, as expected, it's MUCH faster.  I've not put the new version on
clocc yet because it doesn't work with the latest release.
I suppose I should make a version that works the new way if you have
the new clisp and the old way if you don't.  Of course, this won't
do you any good unless you do a cvs build.