On Wed, Sep 02, 2009 at 09:03:12AM -0400, Edward Lewis <ed.le...@neustar.biz> wrote a message of 67 lines which said:
> Without zone transfers, this approach lacks interoperability. The > master and slaves have to be the same make. No, they have "only" to use the same algorithm. One can be in C++ as a PowerDNS back-end and another to be written completely in Perl's Net::DNS. On Wed, Sep 02, 2009 at 05:19:30PM -0400, Edward Lewis <ed.le...@neustar.biz> wrote a message of 58 lines which said: > The WG sputters so much about middleboxes, proxies, etc. that > implement a curbed version of DNS and cry that they are the root of > many ills (UDP fragmentation, capping port 53 traffic to 512 > bytes). Then we turn around and say it's okay to put a hobbled DNS > on port 53. I'm not sure of the exact meaning of "hobbled" but I suspect it is derogatory. You seem to imply that a "custom" DNS server (implementing rules like s/^3ffe:cafe:cafe:\(.*\)$/$1.customer.isp.example/) are always written from scractch, using the bits on the bare metal. Far from it, they typically use a framework allowing the author to just express his/her business rules while the software takes care of the DNS details. Systems like PowerDNS or evldns work that way. Therefore, a server written with them complies with the DNS protocol is certainly not "hobbled" and absolutely not comparable with a broken middlebox. There are deployment issues with these "custom" DNS servers but no ressemblance to the problems exposed in RFC 5625. _______________________________________________ DNSOP mailing list DNSOP@ietf.org https://www.ietf.org/mailman/listinfo/dnsop