In message <AANLkTim22fkSqt__6eFh0k3LDuUtV6orSBoeMFrUADPP@xxxxxxxxxxxxxx>, Phil
lip Hallam-Baker writes:
Sure people can do this, but what is the point?
Let us imagine that as a matter of national policy we are all to live in
smaller houses to save energy. In what respect would such a policy be
realized if everyone bought a second, smaller house and started occupying
both simultaneously?
The critical objective here is to create a situation where people can use
the Internet successfully without requiring a full IPv4 address.
It is not going to be the case that anyone can use pure IPv6 100% of the
time for several decades.
Who is talking pure IPv6 yet? I'm talking multi-homed. Multiple
A's, A's and AAAA's or multiple AAAA's. I don't care which. The
path to one of those addresses is broken and the network is not
reporting it. The question is how to make the application behave
well in such a enviorment and also not be a bad network citizen
when all the addresses are reachable.
Getting people to use IPv6 is not an end in itself. If you have a full IPv4
address there is no point in using IPv6 in parallel.
I will agree that when everything is working as expected it doesn't
matter if you use IPv4 or IPv6. However we are not discussing this
senario. We are discussing the senario where one of the paths is
broken and not returning error messages and you don't know it is
broken without trying it.
The case where this dual stack is going to make sense is where the user has
a shared IPv4 and clean IPv6 and there is a performance advantage to going
through the clean connection.
That's one case but not the only one.
This is not something that I would want to have to code for as an
application programmer. I would want to have a call of the form 'get best
connection for (protocol, domain_name) and have the platform figure it out
using information that I would not want the application programmer to have
access to.
Does it really matter if it in the application or a library. The
problem is still the same.
Mark