Though this page is retained to avoil link missing, but content is quite obsolete. Please pay attention about it.

Benchmark of Authoritative servers

Here's the result of authoritative servers.
You have to understand that this is just a result under particular condition but not the best performance of each implementations.

Objects measured

BIND 9.9.1-P1: set "recursion" as "no"
NSD 3.2.10
Knot DNS v1.0.6

Measurement program

dnsperf 2.0.0.0

Platforms

Name server

CPU: Celeron 2.60GHz
OS: Ubuntu 12.04 LTS
Network interface: Intel Pro/100 (100Base-TX)

Measurement host

CPU: Core i5
OS: FreeBSD 8.3-RELEASE
Network interface: RealTek 8168/8111 B/C/CP/D/DP/E/F PCIe Gigabit Ethernet

What measured

Invoke dnsperf -n 5, 5 times for each implementation and query A RRs of

dummy000-000.example.jp.
        :
        :
dummy255-255.example.jp.

Activate EDNS0 by dnsperf -e to make it practical condition.

Result

qps	BIND	NSD	Knot DNS
Best	4223.9	12009.9	25856.1
Average	4220.4	11991.8	25823.1
Worst	4217.5	11977.7	25742.9

Lost has occured by BIND and NSD. Number of lost was same among each 5 measurement. Knot DNS caused no lost.
Personal consideration on this result is the following. Performance of the measurement system used here is enough to make BIND and NSD to be overdriven, but not for Knot DNS. So Knot DNS may have more capacity and may show more performance if enough load is applied.

lost	BIND	NSD	Knot DNS
%	0.13	0.05	0

Time consumed to invoke and compile.

BIND	NSD(compile)	invoke	Knot DNS(compile)	invoke
0:01.43	0:01.55	0:00.00	0:03.15	0:00.04