[Pdns-users] CPU consumption of pdns_recursor

[Nejedlo, Mark]

We recently replaced some Bind servers with PowerDNS recursor, and were rather surprised to see CPU usage essentially double for the same workload.  My expectation was that the load would be more or less equivalent between the two resolvers.  It looks like the load is centered in the actual pdns_recursor worker threads, not the distributor threads or dnsdist.  Is it expected that CPU usage would be so much higher under PowerDNS?  Is there a debug parameter that can be set to see where PowerDNS is spending its time?

I've already gone through the performance guide and don't see any additional tuning parameters that might help.  We're using LUA, but 1) it's pretty lightweight (a couple hash lookups and modify dq object), and 2) when I disabled the LUA scripting for testing, it didn't make a noticeable difference in the load.

I'm in the process of building a better test env to subject PowerDNS to more controlled testing, but wanted to get the question of what I should be expecting out there to see if anyone could provide some helpful feedback.

Thanks,
Mark

--
XML combines the efficiency of text files with the readability of binary files

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.powerdns.com/pipermail/pdns-users/attachments/20210405/d49489bb/attachment.htm>