performance with many zones

[Jim Reid]

>>>>> "Conny" == Conny  <kop at spice.ascio.net> writes:

    Conny> I run BIND 8.2.2-P5 on Linux 2.2.* systems using i686
    Conny> hardware. Whenever I change a zone and occationally do a
    Conny> reload, the servers naturally stop answering for a short
    Conny> while. When you serve about 25000 individual zones this
    Conny> stop lasts for about 40-60 seconds depending on the
    Conny> hardware.

Well it also depends on the size of the zone files. The more resource
records that have to be loaded by the name server, the longer it takes.

    Conny> Is there any tweaks that can be made that I don't know of
    Conny> to increase performance when having this large amount of
    Conny> zones?

Yes. Use incremental zone reloads with ndc.

    Conny> One funny thing is that it takes just as long before
    Conny> the reload is complete when you've used "ndc reload <zone>
    Conny> ; ndc reload" as it takes doing just an "ndc reload" ..

That's hardly suprising. Both commands you've quoted entail global
reloads of the name server. "ndc reload" (with no arguments) pretty
much makes the name server re-initialise itself, re-read named.conf
and then check all its zones. So a command of "ndc reload <zone>; ndc
reload" is to all intents and purposes just as demanding on the name
server as a command of "ndc reload".

A command of "ndc reload zonename" will make the name server ONLY
reload the zone called zonename. This of couse assumes that you're
using ndc in channel mode to talk to the name server via a UNIX domain
socket. If ndc isn't run in channel mode, it'll just use signals to
communicate with the name server. That means you lose the fine-grained
control to perform per-zone reloading.