[consume-routing] IP addressing idea

[Peter Galbavy]

> No sensible backbone routes RFC1918 addresses, and if your ISP is
> already using them too, you carefully choose a diffent one in the first
> place.

I think we are mixing the term "backbone" too much here. To help qualify my
statements below (which I haven't typed yet) I would refer to three
"backbones":

    1. ISP to connect node to the Internet - by some undefined means
    2. Consume network, somewhat independent of above
    3. Node host company or personal network / backbone

The one that concerns me is 3.

> This wouldn't work except in the simplest of cases. Take me and Steve
> for example. Our nodes overlap. If we are both using the same class C we
> will get conflicts. If you try and allocate a different range of IP
> addresses from the class C to each node, you've limited yourself to a
> total of 254 clients on the entire network.
>
> Under my scheme, Steve uses 10.1.2.x and I use 10.1.1.x - if I wander
> out of the range of my node and over to the far side of his, I simply
> get re-issued a new number by his DHCP server.

My original e-mail did meantion roaming in quotes :) As in "I have no idea"
...

> It's possible we could get away with a block of real addresses bigger
> than a class C, and be careful not to allocate overlapping ranges to
> nodes that are within potential conflict range, but this doesn't give
> you any better protection against misconfiguration, and is considerably
> more difficult to administer.

Ignoring consume for a moment, many people already use RFC1918 addresses on
their own networks, either business or personal, for
NAT/convenience/whatever.

The problem arise not in talking about RFC1918 space over 1 or 2 above but
3. If someone installs a node, is allocated a block from 10/8 etc. but for
some reason installs the node (external politics etc.) behind their own
firewall / NAT box (like me when I have an ISDN dial up doing NAT on
192.168.1/24) then there is a small chance that the two will conflict.

To address Ben's point - shame the word fuck will cause this mail to be
filtered by some; I say fuck 'em :) - it is sometimes not a question of
misconfiguration but simply one of other factors.

OK, all that said, I am a pragmatist; I think using 1918 addressing is
somehow wrong, getting PI space is *easy* - routeing it is hard - but Adam's
comments about overlapping nodes is the one that has made me slow down a
bit. This issue of testing the 802.11 implementation(s) and overlapping
coverage areas means that this needs more thought.

BTW DHCP and roaming ? What lease times were you expecting ? We would need
to keep it to 15 seconds or so - which could get very silly.

Peter