named cleaning interval

[Kevin Darcy]

Mark.Andrews at nominum.com wrote:

> >
> > 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
>
>         If cache cleaning is noticable you most probably don't
>         have enough real memory and the server is paging heavily
>         while the cache is being cleaned and for a short time
>         afterwards as it gets back to a useful working set of
>         pages.
>
>         Mark

Note that it is possible to disable cache cleaning altogether (set the interval
to 0). This might be appropriate for a machine with tons of memory and scarce
CPU resources, and/or if the names being queried are highly clustered, and/or
if the machine is being rebooted frequently for some collateral reason. If
Mark's (memory-exhaustion) hypothesis is correct, however, then disabling cache
cleaning could make your problem even worse than it is now. In that case, the
clearly *best* solution is to buy some more memory, although you might be able
to squeeze out a _little_ performance by tuning your cleaning frequency a
little bit up or down, and/or tuning your max-ncache-ttl setting.

By the way, are you trying to squeeze out some *hypothetical* performance here,
or are your users actually complaining? The clients should have multiple
nameservers in their resolver configurations and fail over relatively quickly
if the first query times out. If you have sophisticated clients, you could even
run caching nameservers on them to reduce their dependence on the problem
machine to a minimum.


- Kevin