[consume-routing] IP addressing idea

[Martin Ling]

On Thu, Jan 11, 2001 at 05:38:47PM +0000, Ben Laurie wrote:
> 
> Martin Ling wrote:
> > 
> > (and now, the whole situation potentially comes round for me again if
> > consume uses 10.* addresses).
> 
> Consume will not route to 10.* addresses. From anywhere.

Okay.

(Do excuse any density on my part here, by the way. I haven't kept up
entirely with the lists, but I've read this thread quite thoroughly)

If this is the case (I assume you mean RFC1918 addresses generally, not
just 10.*), then I don't see any need for the PI address space; *except*
for mobile nodes needing to operate services - which, despite the nice
implications for my wearable, we could do without for now (but add
support for later on a node-by-node basis - nodes will want (and need,
by the 'club' legalities) a say anyway in whether they offer this
service, and to whom).

I also don't see any need to allocate the RFC1918 space, since it will
be completely NATed. If it's thought useful to say 'just use this' to
new nodes, we can give an arbitrary RFC1918 block; we can also append an
instruction to use something else if that interferes with their LAN,
since it doesn't matter.

This, again, possibly comes apart for roaming, but that depends entirely
on the mechanics of 802.11. I'm not clear on these, so I'd appreciate it
if someone could clarify.

If the wireless equipment works like a simple ether, we have exactly the
same issues as for plugging laptops into random networks - these are
solved by DHCP, but you don't move the laptop that much. 15 second DHCP
is indeed silly, but issuing a new DHCP request on noting a loss of
signal (either by non-return of occasional pings or some interface with
the radio hardware) might be more practical in some guise.

However, as I understand it the 802.11 system allows for different
networks in the same radio bandwidth, and that these are distinguished
by some kind of identifiers and the corresponding encryption. Whether
it's possible to 'browse' available networks I don't know.

It may well be that an appropriate roaming solution depends on us
writing some code to detect the need for and specifics of a switch from
one coverage area to another.

Oh, and surely all nodes should be dropping like a brick any traffic
coming in on the wireless interface from or to RFC1918 addresses, and
ignore information advertising them - as such, misconfigured nodes won't
hurt anyone but themselves.


Martin

-- 
-----[ Martin J. Ling ]-----[ http://www.nodezero.org.uk ]-----