Comment re ARP work and ad-hoc networking

[Bruce Simpson]

The ARPv2 snap makes things that much more interesting.

I can foresee that folk may wish to do things e.g. with MANET 
protocols.  In an ad-hoc wireless world, things happen very differently. 
Both ARP and IGMP straddle layer boundaries in the ISO 7 -layer model, 
and are geared towards fixed network topologies which don't change 
dynamically over small t time. Nothing out there in open source land 
really deals with the split all that elegantly.

Because ad-hoc protocols enable the endpoints to discover each other 
dynamically, and there may be multiple ingress/egress points, you can 
effectively populate the ARP table i.e. based on MANET "hello" messages. 
Of course now that ARP is out of the routing table, this probably makes 
things easier in this regard.

But as Sam points out, we may be better off with a new kernel comm 
mechanism. Linux's netlink socket has an Informational RFC, and as such 
is not subject to the GPL -- one cannot copyright an idea.

Whilst implementing it would be a lot of work, it is one good way to 
proceed as it then ties everything together under one API, which would 
greatly help folk writing network apps.

Of course, without a compelling case for going off and doing the work 
(i.e. funded), this is largely hand-waving and just a suggestion I'm 
putting out 'out there'.

just my 2p
BMS