Re: [dnsext] Increasing character limit for registration in internet domain names: 76 or 68 or 91 or 83 or 64 higher the better

[Mark Andrews <[e-mail]>]

In message <[e-mail]>, Paul Vixie writes:
> > From: Florian Weimer <[e-mail]>
> > Date: Tue, 18 Jan 2011 10:44:06 +0000
> > 
> > I wasn't really around when the failure of this experiment was
> > recognized.  Is there a write-up somewhere explaining why this doesn't
> > work as expected?  (Without major protocol surgery, increasing the
> > label length would share a similar fate.)
> 
> no writeup. briefly, it only works if the full end-to-end path knows
> about each new label type. so adding each label type would require 
> incrementing the EDNS version number. this was found to be impractical.

The problem is worse than that.  All the parent server need to be
able to parse the QNAME to be able to issue a referral.  You can't
find the OPT record to do EDNS version processing.

I don't know why we are talking about this at all.   I've see no
evidence that 63 characters is not enough space for any label people
are likely to be typing in even with IDN.

this-is-a-very-long-hostname-that-I-would-never-use-in-practice.isc.org

An entire sentence is encoded in that label.

If you need to encode longer sentences into labels for a database
stored in the DNS then I suggest that you use a cryptographic hash,
base32 encode the result and do collision detection.

Mark
-- 
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: [phone]                 INTERNET: [e-mail]