[MLB-WIRELESS] So how does this routing bit work?

[Ben Anderson]

> On Friday 29 March 2002 11:49 pm, Hamish Moffatt wrote:
> > On Fri, Mar 29, 2002 at 10:41:31PM +0000, vortex wrote:
> > > look. ethernet frames do not route. neither do 802.11b frames.
> >
> > Ethernet can be 'switched.'
>
> Yes. But there are no global swithed ethernet segments - and for good
reasons
> - the underlining distribution mechanism just doesn't scale well. it does
> well what it does best - and that is deliver efficiently in a local wired
> environments.

Sure, that's why I'm suggesting changes to the data-link layer.  Large
changes.  I'm not just talking about blindly 'switching' mac addresses.  I'm
talking about making physical location the switching metric in that layer
until a currently optimal route is discovered, at which point the source
node can source-route all it's packets.


>  The effect is the same as routing - delivery
> > decisions are made based on the source/destination address.
>
> No the effect is not the same. For instance - there is no global frame
route
> distribution for MAC addresses. And if there were, it would be horribly
> inefficient. You think central Internet routers' BGP tables are large
today!
> Just consider the unweidly size of a dynamic MAC routing protocol.
Infeasible.

If it's by physical location, to be able to address a node, and route to it,
there needs to be no information pre-stored other than shortcut information
across the mesh (to allow 'backward steps' towards a good shortcut).
More worrying than the size of the MAC tables is the broadcast war to
generate them!


> > > switching doesn't make sense in 802.11b.
> >
> > Why not? The network is not completely broadcast. There are different
> > channels, and there are directional antennas. In this case, if routing
> > is needed then switching could also be used.
>
> 802.11b has a diferent contention protocol. And I don't see simultaneous
> transmission on multiple wlan channels as switching - it's more like
> multiplexing lines (like MPPP for multiple modems to aggregate bandwidth).

It's listening to, and handing off other peoples traffic in an appropriate
direction that I'm trying to acheive.


> > I'm not saying that it should be used, only that it is technically
> > possible.
>
> Sure. And I agree with you in that it _could_ be used & *is* possible. I'm
> just saying that implementing this will have lasting (detrimental)
> consequences.

With smart decisions, rather than stupid ones, at the data-link layer
(including layer 3 and layer 4 state information in the decisions), then I
think it would have a significantly positive net affect.

Ben.