Performance of auth-only vs recursor.

[Sven Ingebrigt Ulland]

I conducted a simple test within a virtual Xen environment on a host
system with dual AMD Opteron 242 1.6GHz CPUs and 2GB RAM. Two virtual
servers were used, one auth-only and one recursor, both running BIND
9.3.1-2ubuntu1. The auth-only is set up to return an RRset of six IPv4
addresses upon request for the hostname 'www.test.lan' with a TTL of
five hours. The recursor fetches data from the auth-only server.

Now, running queryperf ten consecutive times towards each server,
every time with a list of 10,000 entries containing 'www.test.lan A',
yields the following results:

Auth-only: 11,123 +/- 58 requests/second. (very small variation)
Recursor:   7,541 +/- 18 requests/second. (very small variation)

The recursor was preloaded with the www.test.lan answer by running
'dig @authonly a www.test.lan' before the test.

Now, what I'm wondering is what is the reason for the difference in
performance for the auth-only vs the recursor? The auth-only is
approximately 150% as faster.

regards,
Sven