[ppml] Policy Proposal -- Eliminate Lame Server policy
mysidia at gmail.com
Wed Sep 12 23:37:47 EDT 2007
> This is a rather silly statement. The entire DNS system depends on
> everyone doing things "one right way". The TCP/IP protocol itself
> won't work if everyone did their own version. I could point to numerous
> examples of everyone on the Internet having to do things "one right way"
I consider this reasoning mistaken, because there are differences in
implementation choices, especially when different software is used:
TCP/IP and DNS are not examples of protocols where every
implementation is 100% identical, there can be substantial
differences between any two valid implementations.
DNS, TCP/IP do not rely on everyone doing things "one right way",
in practice, the good implementations have to be flexible and robust enough
to accept the minor deviations that do occur, or could occur in newer
versions of the protocol.
(Deviations such as inclusion of an unknown DNS RR type,
or an unknown TCP extension, different ways of picking outgoing
port numbers, TTLs, etc.)
Protocol standards have optional MAYand SHOULD sections,
in addition to MUST requirements.
Good implementations that follow all MUST requirements of a
well-designed protocol can communicate with other solid implementations
of said protocols that also follow MUST requirements.
It is possible that a implementation of TCP/IP or DNS that misses
some requirements is not sufficiently broken to prevent peers from
There are many desirable practices that are not an absolute requirement
given clearly by any standards documents, in many cases the best
practice may be merely a subtle suggestion in a RFC.
This may well be the case with reverse records -- rfc1034/1035 will
say how these records are formulated, if the reverse exists, but there
is no DNS protocol requirement that every IP in the world has an
existent reverse map (particularly not that addresses of hosts that
don't exist will reverse).
> If you want to argue that in-addr.arpa handing is not a requirement for
> DNS and thus is optional, fine. But arguing to allow everyone to do
> their own thing merely for the sake of being able to do their own thing is
> preposterous on a shared network.
People have the discretion to pick how they interact with the
world; there may be a shared network, but the hosts still belong
to their owner(s).
Provided their choices of implementation are in-line with the relevant
standards, and of good design, they should be able to interoperate on
the shared network.
So long as the parties they need to communicate through/with
can accept their implementation choices.
But if they can't accept those choices, a better reason is needed
than "We didn't come up with the same choice"
More information about the ARIN-PPML