20M req/month seems like a trivial amount that can be handled by a bunch of geographically-distributed VPSes (for failover and better query times) running any s

20M req/month seems like a trivial amount that can be handled by a bunch of geographically-distributed VPSes (for failover and better query times) running any sane (resources-wise) DNS server software. For comparison, we have a bare-metal E3113@3GHz machine that at the time happens to run only a full-fledged PowerDNS server (sqlite + pipe backends, query cache disabled due to split-horizon requirements, and passes queries to a pdns_recursor if can't answer directly). The server easily handles about 500M reqs/month, with negligibly low load average. Judging from top(1) output — pdns_server-instance has about 2-4% CPU consumption, most of which is probably due to disabled caches. But there are probably cheaper options with specialized services.