Re: [Bundled-domain-names] New Version Notification for draft-yao-bundled-name-problem-statement-01.txt

[Stephane Bortzmeyer <[e-mail]>]

On Wed, Aug 24, 2016 at 02:11:33PM +0800,
 Jiankang Yao <[e-mail]> wrote 
 a message of 189 lines which said:

> the following is the new version of PS document.

I've just read it and I have three comments.

One is on the possible size of the "bundle". In french, even if you
limit to the accented characters of the french language, a bundle of
all possible variants for a name with three accented characters can be
on the size of the thousand. It can be larger for other scripts. Is it
a requirment that the solution (whatever it is) work with large
bundles? In the minutes of the Berlin Bar BOF, I see "In scope:
Mapping among any finite number of activated domain names \ Out of
scope:  Infinite number of domain names or variants" which does not
help: 37^63 (number of possible LDH labels) is finite but *huge*.

The second is that you do not mention the solution that people use
today, when the names are in different TLDs: rely on a clever
provisioning system. Just automatically generate the config for all
your domain names. (Hello, DevOps.)

The third is about section 4, application issues. The idea of using at
runtime a DNS request to decide which page to serve seem to me very
bad: latency (there is a reason why Apache does not look up PTR by
default), resiliency (what if the DNS is down), debugging (result of
the HTTP request will depend on a third party).