[dns-operations] latest bind, EDNS & TCP
Simon.Munton at cdns.net
Sat Oct 11 09:21:22 UTC 2014
> Which one(s) have been recently updated and are suspect? Would they really have overwritten previously-configured options, or blithely added new ones which were enabled by default?
Previously, bind has started with bufsize=4096 and reduced it if the
queried server says it can only provide smaller packets.
My understand is that the new behaviour is that it will start with
bufsize=512 and raise it.
But what we are seeing is that some servers are failing to do this and
so making excessive use of TCP.
Although I can't directly this effect with a code change - the timing is
right and its the same area of the code.
And its been a very sudden change (we can identify the day it started
happening) - which suggests code change, as opposed to a config change
(i.e. it happened in a number of places at about the same time).
I said before, the fact remains that servers that are clearly capable of
a larger bufsize but are sticking with a minimum sized one, and so
re-issuing *ALL* their queries over UDP, then immediately again over TCP
(where the target zone is signed).
At a minimum this doubles the workload on the target server, without
taking into account the much higher workload of TCP over UDP.
More information about the dns-operations