Re: LL70: Use of the term "valid"

> From: "Bernard Aboba" <[email protected]>
> Jim Busse said:
>=20
> "Just to clarify my understanding, If the reachability test described =
in
> DNAv4 is skipped, or fails, then IPv4LL can't have a "valid" address,
> simply it will have a "routable" address.  In that event, I may assign =
an
> IPv4LL address additionally on that interface, right?"
>=20
> This might not be a good idea if it's likely that a routable address =
will
> eventually be assigned, since it will cause excessive address changes.
> For example in IEEE 802.11, it could be that the reachability test =
failed
> because of packet loss and a subsequent DHCPDISCOVER would succeed.  =
Is it
> helpful to have an IPv4LL address assigned so as to cause existing
> connections to go down?  Not necessarily.
>=20
> In general, the recommended practice is for the host to only assign an
> IPV4LL address when it is "likely" that no routable address will be
> assigned.  This could occur for example when 802.11 adhoc or Bluetooth =
PAN
> is in use.

Bernard is right - the DNAv4 reachability test is only a rough guide to =
the appropriateness of an address configuration. It can even give a =
false positive - e.g. when I have the address 10.x.y.z configured, I can =
move to another network where the 10/8 range is in use and where the =
reachability test is passed yet my address is not at all appropriate and =
may actually conlict.

As humans we can envisage lots of scenarios and because we can generally =
know something of the state of the whole network, we can usually decide =
quite clearly whether and when it is appropriate to assign an IPv4LL =
address.

From inside the host looking out, it is far less easy to know what is =
going on and come up with an algorithm which does the right thing in all =
cases. What we are trying to achieve is that implementers will =
understand the issues and devise algorithms which will do the right =
thing in the most common cases that they encounter.

The common cases encountered by a PDA with WiFi connectivity are likely =
to be different from a desktop computer with a wired interface, and =
different again from a small embedded device such as a scanner with a =
minimal user interface.

Philip