named cleaning interval

[Julien Mabillard]

Well, I will make some testings with different interval values and monitor
the queries and answers parameters.
I just wanted to know if somebody already experienced that.

I can add that the cleaning concerns about 4000 to 13500 RRs each time. 
I have a master server and a slave one. Both show these results.

thank you



-----Original Message-----
From: Jim Reid [mailto:jim at rfc1035.com]
Sent: Friday, November 24, 2000 2:08 PM
To: Julien Mabillard
Cc: bind-users at isc.org
Subject: Re: named cleaning interval 


>>>>> "Julien" == Julien Mabillard <jum at cw.span.ch> writes:

    Julien> So my question is, for several thousands of hosted zones,
    Julien> what should be the cleaning interval for tuning the
    Julien> performance of the bind server?

It depends. The default cleaning interval should be "good enough" for
most name servers. Whether its good enough for you depends on things
you've not told us (and probably don't know yourself). For instance,
how long (on average) it takes to clean the cache; how much of the
name server's data structure is cached RRs and how much is used by the
RRs loaded off zone files; what percentage of the traffic is queries
for names the server can answer authoritatively and what percentage
involves your name server querying other servers; how many system
queries does your server make; what fraction of queries come from
other name servers and what come from stub resolvers; etc, etc.

If lots of people are noticing that the server isn't responding
because cache cleaning takes too long, you probably should use a
longer cleaning interval. Hopefully this would mean the cleaning
(garbage collection actually) took place at times when your users
weren't querying the server. OTOH, if the cleaning interval was
shorter, it's possible that the time taken to do the clean up would be
lower because each run had fewer expired RRs to get rid of and
therefore had less work to do.

Personally, I doubt if anyone will notice the cache cleaning outages.
And if they did, there are better ways to solve the problem. One would
be to separate the name servers used to serve the zones to other name
servers from those used by stub resolvers to make queries. Another
would be to run a threaded name server such as BIND9 which can walk
and chew gum at the same time.

Why don't you experiment with different cleaning intervals and report
your findings here?