[Cu-wireless] MIT Roofnet article at TechnologyReview.com

[David Young]

On Tue, Sep 02, 2003 at 01:58:52PM -0500, Sascha Meinrath wrote:
> Just a quick follow-up -- the software is open source and available for 
> download at: http://www.pdos.lcs.mit.edu/grid/software.html 
> 
> It seems to me that we may want to have one of our Linux-savvy folks take 
> a look at it.  I would also love to know if there are solutions that can 
> be implemented into our own software or whether we should have folks 
> working on a separate fork (if people are interested).  
> 
> Thoughts?

AFAICT, the most important research outcome from the MIT Roofnet is ETX,
a "routing metric." A routing metric is a way of assigning "scores" or
"costs" to the links in a network in order to bias the routing algorithm's
choice of paths. The ETX algorithm is simple, and it is equally applicable
to an OSPF network as to the DSR-hybrid which Roofnet uses. Essentially,
you weight links in the network using the formula

       1
   ---------
    df * dr

where df is the proportion (between 0 and 1) of ten 802.11 broadcast
packets that were received over the link in the "forward direction" in
the last ten seconds, and dr is the proportion received in the "reverse"
direction. It is a little more complicated. IIRC, they use an exponential
weighted average of 1 / (df * dr) over time.

I am not expert enough at OSPF to know how easily we can use ETX on our
network, especially not in the point-to-multipoint mode. Maybe Stephane
can comment? AFAIK, the only thing that OSPF lets us assign costs to is an
"interface." Am I wrong?

Incidentally, the benefits of ETX on our *current* network will be
nil. However, with "fill in," we will definitely need ETX.

MIT uses ETX with a very complicated routing protocol which they call
SrcRR. It is a variant of Dynamic Source Routing (DSR). You can read more
about it on their web page. I think CUW will do better by using ETX with
a routing protocol such as Hazy-Sighted Link State (HSLS). HSLS admits
a *much* more parsimonious implementation, and it is predicted to scale
better than SrcRR.

The work-conserving thing to do is to continue with the CUW software,
adding ETX to our OSPF routing, or producing an HSLS daemon with ETX.

Dave

-- 
David Young             OJC Technologies
dyoung at ojctech.com      Urbana, IL * (217) 278-3933