To:
dnsop@cafax.se
From:
Rob Austein <sra+dnsop@hactrn.net>
Date:
Mon, 24 Jun 2002 18:30:40 -0400
In-Reply-To:
<a05111b03b93d41f48c5a@[208.58.208.43]>
Sender:
owner-dnsop@cafax.se
User-Agent:
Wanderlust/2.4.1 (Stand By Me) SEMI/1.13.7 (Awazu) FLIM/1.13.2 (Kasanui) Emacs/20.7 (i386--freebsd) MULE/4.0 (HANANOEN)
Subject:
Re: is this proper behavior?
At Mon, 24 Jun 2002 17:40:05 -0400, Edward Lewis wrote: > > (This is the question that led me to ask this on DNSOP.) Is it > acceptable for a server to answer for some zones and not at all for > others? Wouldn't this cause a problem for DNS implementations that > measure round trip times in order to select one server over another? Depends on the algorithm. If the observed behavior were that the name server repeatedly answers queries for different names at different speeds (that is, if there were a repeatable pattern to the round trip times), I'd agree that it was a measurement problem. As you describe it, though, it's a sampling problem (which data does one feed into the RTT estimator?), which is a subtly different thing. TCP, for example, would not regard lack of response as any kind of sample at all (that is, the connection would of course eventually time out, and the RTO would back off, but the RTT estimate would not change -- see RFC 1122 4.2.3.1 and follow the references to the SIGCOMM papers). With all that said, the behavior you describe is questionable in light of the robustness principle, so it's probably a bad thing unless there's a very good reason for doing things that way.