Questions about DNS performance

Joaquin J. Domens jdomens at corp.terra.es
Tue Feb 18 10:23:37 UTC 2003


Hi all,
I work for a Spanish ISP and we're experimenting some problems at the
peek hour (18:00 european time) as I get all my servers at 90%  of cpu
usage.

I would like to know wheter this is normal or not.

We're running BIND 8.3.4 on 4 solaris boxes (NetraT1) with 512 MB RAM
each one. DNS is splitted so this boxes are without any authoritative
zones, they running just BIND.

This boxes are under a LB with round robin blancing.

The statistics during the day are something like this ....

#netstat -I hme1 1                =20

    input   hme1      output           input  (Total)    output
packets errs  packets errs  colls  packets errs  packets errs  colls=20
563     0     590     0     0      565     0     591     0     0    =20
521     0     719     0     0      523     0     721     0     0    =20
509     0     550     0     0      511     0     552     0     0    =20
484     0     588     0     0      494     0     598     0     0    =20
564     0     637     0     0      566     0     639     0     0    =20
549     0     606     0     0      551     0     608     0     0    =20
543     0     665     0     0      546     0     668     0     0    =20
547     0     637     0     0      549     0     639     0     0    =20
545     0     555     0     0      547     0     557     0     0    =20
535     0     593     0     0      537     0     595     0     0    =20
648     0     779     0     0      650     0     781     0     0    =20
549     0     688     0     0      551     0     690     0     0    =20

And with this output of the interface the CPU load it's about  60% ....

 PID USERNAME THR PRI NICE  SIZE   RES STATE    TIME    CPU COMMAND
 24909 root       1  30    0  295M  291M run     75.9H 61.18% named

With this escenario we're about 90% of CPU at the peak hour.
In the LB output traffic in the peak hour it's about 6 MB of bandwidth.

Do yuo guys think this is a normal behaviour for the configuration
showed or you think It could be due some misconfigurations ?

I mean, I think they are serving according to the traffic, maybe we will
need more RAM or boxes to serve it correctly, but I don't know what a
system with this kind of load should be running.

Is anyone running BIND in large systems as here ?

TIA


--
--------------------------------------------------
Joaquin J. Domens
=C1rea de Tecnolog=EDa
Departamento de Producci=F3n / Aplicaciones
--------------------------------------------------
Terra Networks Espa=F1a S.A.
Via dos Castillas 7, Pozuelo de Alarc=F3n
Madrid, Espa=F1a
Tel. (34) 91-452-33-71
joaquin.domens at corp.terra.es
--------------------------------------------------
Mercado Continuo: TRR |  Nasdaq: TRLY
--------------------------------------------------
http://www.terra.es
--------------------------------------------------



More information about the bind-users mailing list