If memory serves me right, Dinesh Nair wrote:
> On 11/23/05 00:35 Bruce A. Mah said the following:
> > similar to m0n0bsd in that it has a compressed MFS root filesystem, and
> > therefore doesn't require the boot media after the kernel is loaded.
> 
> exactly, and the benefit (to me) of this is that the boxen can be powered 
> off wantonly without worrying about a filesystem glitch if fsck doesnt do 
> its thing on a bootup.

I know.  nanobsd mounts all of its filesystems read-only as well.  See:

http://www.freebsd.org/projects/nanobsd/

> > because I don't have to deal with compressing and recompressing
> > filesystem images to make a change.
> 
> it's not too bad if it's scripted. additionally, i usually pxe boot a test 
> m0n0wall image during development just so i dont have to keep writing to CF 
> cards.

This still means you need to reboot, but OK.  I didn't try the PXE thing
when I was doing the filtered bridging stuff and I found this extremely
difficult to work with (on the other hand, expensive edit-compile-debug
cycles forced me to write better code).

> > downloads through my net4801 (filtering bridge with a few dozen PF
> > rules) can saturate my 6 Mbps DSL downlink.  This is of course not a
> 
> which should be enough for most xDSL users, however those using m0n0wall in 
> pure LAN environments may need to get as close to line speed as possible.

Absolutely.  I said that this wasn't much of a test.  :-)

One old datapoint:  I measured about 25 Mbps between a couple of my
machines going through the bridge plus a 100 Mbps switch (endhosts were
400 MHz P-II and 1 GHz P-III, both RELENG_5; bridge was a net4801
running HEAD from about August).  It was 86 Mbps without the bridge.  I
did not try any particular optimizations such as interface polling but I
did disable INVARIANTS and WITNESS.

I haven't had a chance to repeat this experiment with 6.0-STABLE but I
know there was some work in the network stack since I did this
experiment.

Cheers,

Bruce.