Welcome to my little side project. It has taken
me a while to get around to finding the time to
put this back in.. whew.

What we have here currently is some basic changes that were
done at Adara Networks to the way the lower level stack works
with device drivers. Basically all mbufs have a priority
field (cosqos), this field gets set to the highest number
by default (usually queue 7 by default if we have
8 queues defined 0-7). Now whenever a driver goes
to send out data, it starts at queue 0, and works
its way out. This means that *yes* you can get
packet starvation of things placed in queue 7 while
other things are going to queue 0.

There are also some new methods added to if_net that
deal with managment of the queues/rings. This is
the start of where I plan on putting an AQM module
in place. Right now its just drop-tail if the queue
gets full, at some point I will try to add in both
codel and possibly Cisco's Pi. Pi will need to be a
klm since C has patents on it (the typical C patent
of you can have a free license unless you sue me).
Codel is open created by Van Jacobson and Kathy Nichols
only it was done in Linux first, so I will have to 
read the latest RFC's and try to do a clean implementation
of that so we have no GNU contamination (this may take
a while depending on how much time I can pull away
from Adara, or if I can justify doing it as an
Adara project). 

Now once I have all this settled in to head and know it
all compiles (not sure since I was merging old code from
quite a while ago that this branch yet compiles :-D), I
want to move on and make the IP packet processing stack
lock-less. What I mean by that is I would like to see
*no* locks in the packet forwarding path. Only when you
change things (radix trees etc) will locks be applied.
How I will do this is by adding a garbage collector entity
that "defers" removal of things long enough for packets
being forwarded (during a change) to be out of the system
*before* removal of the old entity. This sounds a bit
strange but it works quite well as long as one can
bound the time period needed before the garbage
collector gets active.

This will come later assuming I get time.. of course
my life as been full of ENOTIME errors, but hopefully
is slowing down here a bit :-)

You are welcome to play in here and contribute. You can
email me at rrs at freebsd.org (or rrs at lakerest.net.. thats
where the mail goes anyway :D)

Best wishes
Randall Stewart