Re: DNSEXT WG Last Call: Message Size

["D. J. Bernstein" <djb@cr.yp.to>]

I object to this document, for several reasons:

   * Sections 1.2 and 1.3 are silly. Normal A6 responses don't overflow
     a 512-byte UDP packet. Root servers don't need IPv6 addresses. The
     claimed 13-server limit for UDP packets can be shattered with no
     protocol changes.

   * It's a layering violation to say ``MUST use EDNS0'' when you mean
     ``MUST use a protocol that handles 1024-byte packets efficiently.''
     Why shouldn't we use DNS over, for example, an improved version of
     T/TCP that can handle 1024-byte packets as efficiently as UDP?

   * The underlying efficiency argument is bogus. The document claims
     without justification that requiring TCP ``will cause significant
     overhead and delays.'' In fact, unless there are _frequent_ TCP
     packets, the extra load from those packets is a negligible portion
     of DNS load, not to mention total load.

Does anyone have logs showing a noticeable number of TCP retries? What
was the actual server load, and what difference did TCP make? What are
some examples of the responses that forced TCP retries?

---Dan

P.S. I would like to receive Cc's of all further messages on this topic,
for reasons explained in http://cr.yp.to/djbdns/namedroppers.html.


to unsubscribe send a message to namedroppers-request@ops.ietf.org with
the word 'unsubscribe' in a single line as the message text body.