hi,

> This paper was presented at the 2001 ACM SIGCOMM Internet Measurement
> Workshop.  I went to the 2002 one last November (in Marseilles) and
> there were many excellent papers there too.  The proceedings,
> including full PDF of all papers, can be found here:
> 
> http://www.icir.org/vern/imw-2002/proceedings.html
> 
> Some papers are a bit on the theoretical side, but many are actually
> quite interesting for network operators.  There were two papers in the
> "DNS" session, but only the second one really talks about DNS
> performance observations.

Thanks, this will keep me busy for quite a while :)

> What often causes intermittent problems like this is when one of the
> alternative delegation paths to the name is misconfigured,
> i.e. lame-delegated or so.  Then you can resolve the name when you
> happen to use the "good" path, but after a timeout you may use the
> "bad" path and encounter a SERVFAIL somewhere (a timeout would be
> harmless).
> 
> This is just one reason why you should be very careful to keep your
> delegations up to date, monitor all your secondaries for proper
> resolution of your names and proper zone updates, and ALWAYS inform
> your authoritative name server (slave or other) operators when you
> move a zone AWAY from them, and make sure the zone is removed from the
> old server(s).

Both DNS servers are operated by us, geographically diverted, rsync
mirrored at an intervall of 5 minutes. If my understanding is right,
a lame delegation ..

  .. pointing to an existing DNS server, that does not know your
  zone, results in NXDOMAIN, which will/can cause negative caching.
  Your zone will be down for (?) time (who decides the TTL for
  negative caching?).

  .. pointing to a NON-existing DNS server results in SERVFAIL,
  which SHOULD cause the dns resolver to pick the other server -- the
  "good path". Otherwise the DNS architecture would really not make big
  sense.

Or am I mistaken? It is so terribly hard to trace problems within the
DNS architecture. I first suspected a too low concurreny limit, but
a manual inspection showed not more than 1 request / second so this
does not seem to be the cause.

However, I fixed the problem so this is no big deal, but still, I'm
interested in the inner workings of the DNS system.

-daniel
----------------------------------------------
[EMAIL PROTECTED] Maillist-Archive:
http://www.mail-archive.com/swinog%40swinog.ch/

Reply via email to