[65480] in North American Network Operators' Group
RE: Above.net problems ??
daemon@ATHENA.MIT.EDU (Duane Wessels)
Wed Nov 26 17:00:20 2003
Date: Wed, 26 Nov 2003 11:31:32 -0700 (MST)
From: Duane Wessels <cee4@packet-pushers.com>
To: "'nanog@merit.edu'" <nanog@merit.edu>
Cc: Arjan Hulsebos <ahulsebos@corp.home.nl>
In-Reply-To: <53B03DA73719D5119E6A0008C71E1D7804DD9660@svr-mail1.corp.home.nl>
Errors-To: owner-nanog-outgoing@merit.edu
> You'd think that an unresponsive nameserver would be flagged dead, and such
> information be cached. Does anyone know whether that's actually done in Bind
> 8.3.4? Or perhaps not by default?
This certainly does not happen when all authoritative nameservers
are unresponsive. See http://www.nanog.org/mtg-0310/wessels.html,
in particular pages 23 and 24 of the slides.
In my simulations with 100% packet loss, DNS caches running BIND8,
dnscache, W2000, and W2003 all amplified the user's query rates.
Only BIND9 attenuated.
The results do depend on the actual query rate, however. At a
higher query rate, the other caches would/should attenuate as well
(perhaps reaching their hard-coded rate limits), but I don't have
the exact numbers.
It would be interesting to repeat the simulation and take out, say,
half of a set of authoritative nameservers during the middle of the
test.
Duane W.
