Sure, but was a silly mistake I made, Since I was testing on localhost over to another IP on the same machine, all the queries where going from the Local IP to loopback (even though I forced on the remotes flag w/ "via"). So all I had to do was actually serve everything through loopback (only different ports) to make things work. The debug showed the incoming query via *127.0.0.1 <http://127.0.0.1>:<randomport>* The wrong setup I made was: Knotd listening on 192.168.0.1:53; PowerDNS listening on 127.0.0.1:53; PowerDNS couldn't talk to Knotd there, so I changed to: Knotd listening on 127.0.0.1:*53*; PowerDNS listening on 127.0.0.1:*54*; And then they started talking AXFR properly Since this is only a test setup to a larger scale migration I was just testing how AXFR performs w/ tons of data to load (250mb of zone files) and frequent updates that I was running in the same virtual machine. BTW, is a good practice to use the zones over tmpfs or since the knotd already throw everything at RAM on the runtime I don't need it? I didn't liked the load times (but I do understand the existence of knotc and why I should not keep "restarting" knotd). On Tue, Jul 7, 2015 at 4:34 PM, Ondřej Surý &lt;ondrej.sury(a)nic.cz&gt; wrote: -- [ ]'s Filipe Cifali Stangler