>[Chris (Torek), how about you?  Any ideas?]

Not really, no.  I know kind of what it should do and sort-of-vaguely
how Mike planned to do it, but that is about it.

>Connect thru STREAMS is also less than optimal.  Pushing all the required
>modules atop the STREAM head costs some time.

It *should* be unnoticeable.  The problem with SysV STREAMS, if you
have ever looked at it, is that it bends the stream model (a simple
up- and downward dataflow) all out of shape to accomodate:

	- control messages (ioctl et al), with timeouts
	  (this simply does not fit in the model! I do not know
	  of any good way to do it)
	- multiplexors

The stream multiplexors in particular are super-hairy, quite literally
thousands of lines of code to do what the IP protocol switch or the
TCP control blocks do in hundreds.  There has got to be a better way.

Chris