On Mon, 20 May 1996, der Mouse wrote:

> Personally, I'd just tell the serial-IP encoding program to dump a
> traffic sample, but given the all-in-the-kernel implementation of the
> stock NetBSD slip and ppp (which I assume you're using), that isn't
> really possible.

I've heard other noises about our all-in-the-kernel vs. a userland
PPP lately (to do with packet filtering and demand dialing, I
believe), and why a userland PPP would be nice. However, before
anyone gets any ideas :-) I'd like to point out that this will blow
up on slower machines. My Sun 3/60, for example, uses a good chuck
of its CPU (a 20 MHz 68020) dealing with PPP when data is coming
in full speed at 38,400 bps. Having to cross the kernel/user barrier
for each packet as well might well kill it.

Aside from the fact that some of us just like our Sun 3s and don't
want to upgrade, I think these slower machines provide a good
platform for encouraging optimisation of the OS. You can get away
with a lot of sloppiness on a Pentium and never notice it, but on
a Sun 3 you notice.

BTW, if anybody's working on kernel profiling code, I'd be interested
in hearing about it. One area that needs work is filesystem I/O,
from the looks of comparisons between my 486s and Pentiums.

cjs

Curt Sampson    curt@portal.ca		Info at http://www.portal.ca/
Internet Portal Services, Inc.	
Vancouver, BC   (604) 257-9400		De gustibus, aut bene aut nihil.