Blue wrote: > Hi, all: > > Currently a dual-stack device could get both IPv4 and IPv6 DNS server > list from either manually configuration or DHCPv4/v6. Is there any > regulation about the querying sequence when a DNS client wants to invoke > a query? I have dug into the RFC database and could not find any > documents concerning about the question. Or it's just an implementation > issue?
Like any IPv4 -> IPv6 transition mechanism: first try IPv6, then try IPv4.
At least that is how the RFC's specify it (don't ask me which one, but
one of them does ;)
There is one problem with this though: lets take a dual stacked host A,
it has IPv4 and IPv6 connectivity, but IPv6 is broken, the host wants to
go to a dual-stacked webserver, it tries to connect over IPv6, it times
out (no icmp coming back for whatever reason), and tries IPv4. Thus it
does reach the server, but after a timeout. If you thus have a user,
they will percieve it as a crappy network and might go somewhere else.
As such, depending on the application and where it will be deployed it
might be smarter to have an option of enabling IPv6 later. IMHO it would
still have been good to have a OS/user-global way to specify this. Some
dns-resolver libraries support this actually.
The real solution to this problem is of course to simply turn of IPv6 or
to fix the network ("ip -6 ro add <prefix>/32 dev lo" helps ;). Of
course it might also happen the other way around: broken IPv4 but
working IPv6. But at the moment there are IPv6 'networks' which are very
crappily maintained and then those circumstances happen, the user will
then only say one thing: IPv6 sucks... as that is how they perceive it.
Can't blame them for that now can we. When thus having an IPv6 enabled
network, also make sure one actually manages it, the same as IPv4.
Note that several browsers have for that exact reason an option to have
IPv4 go first and then do IPv6 later. The default Mac OS X browser does
that for instance, and Opera also has afaik at least the option.
DNS uses round-robin and server selection in quite some implementations
anyway, as such in the case of DNS this should not give a big problem.
Giving the user the option to choose is always a good thing.
Greets,
Jeroen
signature.asc
Description: OpenPGP digital signature
