global forwarders - current BIND9 behaviour documentation

[ip admin]

Hi,

anybody there who can provide a definitive answer on the current BIND 9.7
(or higher) global forwarder behaviour?

I did find the following info before on using multiple forwarders:

https://lists.isc.org/pipermail/bind-users/2007-September/067830.html

My expectation based on that is that the fastest responding forwarder will
basically always be used until a timeout may occur, i.e. when specifying
three forwarders one will be the prefered one based on SRTT and the others
are only used if the prefered one goes down.

First of all when doing 'rndc dumpdb -all' I cannot find my forwarders' IP
addresses in the named_dump.db at all as explained in the posting above
(BIND 9.7.3-P3 on Linux), so I cannot verify the SRTTs. 'rndc stats' /
named.stats does not show any info on the forwarders as well.

Also by doing a tcpdump I can see that all three forwarders I have
specified are constantly used. However it is not a real round-robin but
roughly a 3:2:1 ratio instead (i.e. one receives approx 3 times the number
of queries compared to the third one, the other one receives 2 times the
number of queries compared to the 3rd one). In fact the 3:2:1 distribution
reflects the response time I can manually determine by running dig against
all forwarders - the one which responds quickest gets the most queries and
the one which is slowest gets the fewest queries.

My server receives quite a few queries (approx 10.000 within a minute). Any
idea if the DNS-Server will send every 10th query or so the slower
forwarders?

I also tried to set the logging level to debug 10 for category resolver but
no luck at all in finding out which forwarder is used (and why).

So . . . if somebody could explain what the current behaviour is supposed
to be that would be helpful.

Regards
 Tom
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.isc.org/pipermail/bind-users/attachments/20120725/925163ab/attachment.html>