Re: LL32 Multihoming
Robert Elz <[email protected]>
| Newsgroups | gmane.ietf.zeroconf |
|---|---|
| Message-ID | <[email protected]> |
Date: Wed, 2 Jul 2003 11:47:45 +0100
From: "Philip Nye" <[email protected]>
Message-ID: <002f01c34087$60068af0$131010ac@aldebaran>
| The use of "any" implies that two interfaces on the same host and connected
| to the same link would not recognise each others ARPs as conflicts. This is
| not independence.
No, but nor does anything claim that the interfaces are independent.
What it says is to run the algorithm independently on each interface.
That is: the order in which the two happen (including in parallel)
should be irrelevant, and the results of running the algorithm on the two
should in no way depend upon each other. That's all it means.
It in no way precludes the algorithm (which is to be run twice, or more,
independently) from examining all kinds of information about anything
that happens to be in the system, including all of its interfaces.
| The use of "host's own IP address" is also ambiguous when the host plainly
| has multiple IP addresses.
Not really ambiguous, but perhaps that could be extended to say "any of the
host's IP addresses on the interface". It is OK as it is, it doesn't
actually say that there can only be one - eg: if I own 4 cars, each one
of those cars "my own car", even though it isn't my only car.
| For independence, each interface-IP address pair
| needs to be treated independently.
No-one wants independence, all that is wanted is that the algorithm be
run independently. Read the text...
| It is already clear that LL addressing will not work on multiple interfaces
| on the same host unless traffic is closely coupled with specific interfaces
| in ways which common APIs have tried to avoid. This is another manifestation
| of that effect.
No, it isn't. This has nothing whatever to do with the APIs. Assigning
addresses on multiple interfaces (which is what this is concerned with)
is trivial. It is only using the things once assigned where difficulties
start to occur, and only because the interface between the stack and the
application provides no easy way to be selective about which link is
intended (the IPv6 API doesn't have such issues).
kre