truncated responses vs. minimal-responses?

Mike Hoskins (michoski) michoski at cisco.com
Tue Nov 27 17:41:42 UTC 2012


-----Original Message-----

From: Matus UHLAR - fantomas <uhlar at fantomas.sk>
Date: Tuesday, November 27, 2012 12:28 PM
To: "bind-users at lists.isc.org" <bind-users at lists.isc.org>
Subject: truncated responses vs. minimal-responses?

>Hello,
>
>last few weeks I have seen many discussions over UDP truncating and using
>"minimal-responses yes;" to prevent BIDN from doing that.
>
>I've read article stating that nameserver should avoid truncating packets
>even by skipping additional and authority sections in its responses, which
>should mean that using minimal-responses would not help.
>
>However, I've seen a few mails mentioning that a query can get truncated
>when the authority section is too big and advices to turn
>minimal-responses
>on.
>
>Reading the 9.9.2 docs and even looking at the sources (I am not a C
>coder)
>did not help me with this.

It seems it should help...  less bits in the packet relating to additional
and authority should leave room for other data.

That said, I think the better way (when possible) is to adjust RRs not to
return "too much data" (e.g. NS, A, etc. not returning more than ~8 hosts
-- which in turn could be multicast, load balanced, etc to get the desired
scale).

Akamai, for example, defaults to limiting up to 8 "RDATAs" per RR (or
however you'd describe that).  If you add 20 As for a name you'll rotate
through 8 at a time.  You can request more at your own risk...they assume
you'll ensure the larger answer will fit in a UDP packet and not cause TCP
responses which cripple performance.




More information about the bind-users mailing list