[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: dual stack & IPv6 on by default

On Sunday, March 16, 2003, at 10:38 AM, Erik Nordmark wrote:
That's a really good question. I'm not sure what would be the best way
to address this. What we're saying in this draft is that operational
experience has shown that this part of ND could be problematic in some
circumstances. Does ND need to be changed as a result? Probably not.
There are no MUST's surrounding that bit of ND. As implementors, we're
free to interpret the spec in a sane mannor. With this draft, we're
pointing out reasons why implementors may want to be careful with that
part of ND. There is informational value in this alone without having
to muck with the ND spec itself.
A possible way to approach this problem would be to make
the choice between A and AAAA be a function of whether there
is one or more IPv6 route off-link (or at least one IPv6 router sending RAs).

That way you don't have to tweak ND. But you do end up with some
having e.g. getaddrinfo() check with the kernel. I think getaddrinfo()
needs to already do some checks in order to implement the default address
selection document.
What is the benefit of doing so versus doing the change in ndpd?






If you don't have a route to the destination, why try to reach it
on-link just in case the destination might happen to be on-link?  Is
there a situation where this would be useful?



If you have two nodes communicating on a link and the router
dies for long enough time it seems useful to be able to
continue to communicate.



In that case, it make sense to maintain the local subnet route
but remove the default route. Destination on the local subnet will
still be reachable, other further destination won't and we will not
try them. This can be done in ndpd.





Of course, the communication will fail
once the addresses become invalid but that might take a long time.

  Erik