To:
Brad Knowles <brad.knowles@skynet.be>
Cc:
Bruce Campbell <bruce.campbell@ripe.net>, Randy Bush <randy@psg.com>, <dnsop@cafax.se>
From:
David Conrad <david.conrad@nominum.com>
Date:
Tue, 29 Apr 2003 17:33:57 -0700
In-Reply-To:
<a0521064bbad402d06796@[10.0.1.2]>
Sender:
owner-dnsop@cafax.se
Subject:
Re: draft-ietf-dnsop-serverid-01.txt
Brad, On Tuesday, April 29, 2003, at 03:24 AM, Brad Knowles wrote: > Still, it's not deterministic, and any number of samples that can be > counted by a human being are going to be vanishingly small compared to > the real workload of the server. At best, it's a "good" indication, > for some value of "good". Unless the DNS protocol is modified to include server identifying information for each and every query (and we modify DNS using applications to archive that information), there is no solution to this problem that I am aware of. Regardless of whether anycast is being used, you first have to notice a problem, then you have to figure out which IP address provided the bad response (since ordering of NSes between queries is not guaranteed). Given this is a multi-step process, the use of sampling is pretty much required. However, this whole issue is a rare corner case. If the draft was talking about something people had to rely on, the lack of determinism in a corner case might be worth arguing about. However, the draft isn't. It is talking about a minor modification to an existing convention that merely _helps_ when trying to track down a problem. Rgds, -drc #---------------------------------------------------------------------- # To unsubscribe, send a message to <dnsop-request@cafax.se>.