Policy Proposal 2001-3

[Alec H. Peterson]

----- Original Message -----
From: "Randy Bush" <randy at psg.com>
To: "Alec H. Peterson" <ahp at hilander.com>
Cc: <v6wg at arin.net>; <ppml at arin.net>
Sent: Wednesday, September 26, 2001 14:05
Subject: Re: Policy Proposal 2001-3


>
> alex,

Since you can't even get my name right I suspect it is you who is having the
literacy problem.

> i suspect that you have a problem with your mail user agent.  i
> never said any such thing, at least not in today's mail.  i just grepped
> it.  other things also make me suspect you have a problem with your
> mail reading agent.

Perhaps 'working fine' was not an accurate paraphrasing, however:

[snip]
and how does being multi-homed mean that one will not time out on the server
if the deamon is dead but the box is up and reachable?

and do note that the long timeout is not relevant if the box is not
reachable.
[snip]

Which to my mind was your assertion that there are some things PI space won't
fix.  You will also note I did not dispute your assertions, but rather pointed
out that they were beyond the scope of the discussion.

Just to remind everybody (and you Randy since apparently you have forgotten)
the discussion at hand is whether to modify ARIN allocation policy to allow
critical DNS server (gTLD, ccTLD, etc) to get small, PI blocks of address space
directly from ARIN.  From my standpoint, the main reason to do this is to allow
such servers to multi-home effectively, to improve reliability.

Randy has asserted that is not necessary, since DNS timeouts accomplish the
same thing with respect to reliablity.  I would agree, if we changed it to mean
'eventual reliability'.  However, I personally feel we can do better than
eventual reliability, and since we can we very well should.  Allowing critical
DNS servers to multi-home makes it so it is less likely that clients querying
these servers will need to resort to waiting for queries to timeout.  Instead,
those queries are more likely to be answered on the first try, since the
network connectivity for the server can be provided by multiple providers
easily.

But then, I must be wrong, because the RFCs say so, and we know they're always
right.

Alec