named process increasing dramatically

Mathias Körber mathias at koerber.org
Sun Dec 10 01:49:15 UTC 2000


> today I notice that when i use the top command on my dns server (bind
> 8.2.2P5 under FreeBSD 3.4), my named process reach almost 80-90 % of =
CPU.

How often does this happen, for how long? How large is your cache (try =
ps or
dump the cache to find out)? Ho wmany zones does your server serve, =
and/or
is it used by many clients for lookups?

My first guess would be that this is a routine maintenance, where BIND =
goes
through all its data and removes expired records etc. This usually =
happens
every hour or so. If the cache has grown large, this can take a while =
and
during this time bind is very busy.

Also, if your named size has grown so big it gets swapped/pages out, =
bind
will be overly busy paging in all required pages (and during maintenance =
it
access *all* data, so all pages will be pages in at some time). BIND =
really
does require that all its data fit into RAM (and actually stay there)..

To find out whether parts of your bind-process have been swapped out,
do a ps(1) on it and compare the process size with the RSS 9resident set =
size).
If the RSS is markedly lower than the total process size, the difference
has been pages out.


Other possible resons: someone is hammering your server with queries, =
you should
be able to ascertain that by turning on query logging or debugging. =
(debugging would
also show maintenance activity!)

HTH HAND
>=20
> link to the internet is currently very low
>=20
> the process increment cause very slow responses from the server on dns
> queries
>=20
> i haven't found anything meaningful or unusual from the log =
messages...
>=20
> what could possibly happening on my server ?




More information about the bind-users mailing list