benchmark question

[Dario B. Bestetti (OpServices)]

----- "David Dumortier" <david.dumortier at linagora.com> escreveu:

> Hi Patrick, hi all,
> 
> Dario B. Bestetti (OpServices) a écrit :
> >
> >
> > ----- "Patrick Morris" <patrick.morris at hp.com> escreveu:
> >
> >> shadih rahman wrote:
> >> > List,
> >> >    Is there any information as to what is the maximum number of
> >> > services one can run on a single box with single instance of
> Nagios?
> >>
> >> > I did not find any concrete data on this.
> >> >
> >> > I am running a single instance of nagios on a quad core 2.5 GHZ
> >> > machine with with 4 Gigs of RAM.
> >> >
> >> > I have total of 7359 service check running on this box.  I have
> also
> >>
> >> > ndoutils running on the same box as backend.
> >> >
> >> > My total service checks is going to five fold very soon with a
> lot
> >> of
> >> > nrpe checks.
> >> >
> >> > Now, my question is should I run multiple instance of nagios on
> the
> >>
> >> > same box or a single instance will be able to handle about 30000
> >> > service checks?
> >>
> >> In my experience, you'll run into issues right around 8,000 checks
> or
> >>
> >> so, depending no output verbosity, on a fairly stock setup. This
> is
> >> due
> >> to the size of the pipe used to temporarily store check results,
> at
> >> least on Linux (and you may be seeing it already with your number
> of
> >> checks). Around that point, even with quite a bit of tuning, the
> check
> >>
> >> results will fill the pipe in less than a second, which is the
> minimum
> >>
> >> amount of time I've been able to configure Nagios to flush it.
> When
> >> that
> >> happens, latencies go through the roof.
> >>
> >> Distributing the checks doesn't solve the problem if you're still
> >> sending the results to a centralized Nagios instance, since one
> >> machine
> >> still needs to process all of them.
> >>
> >> I'm working through this situation now, and it's looking like it
> may
> >> take a custom kernel with a larger pipe size to handle
> >> it.
> >>
> >>
> > Shadih, we have a customer running over 20.000 service checks in a
> single
> > box.
> >
> > The box is 2x Intel Dual-core Xeon 3Ghz with 8Gb RAM. The latency
> is
> > around 0.7s. The checks are performed between 3min and 5min. They
> have
> > around 40 simultaneous users.
> >
> > _________________________________________________
> > Dario B. Bestetti        OpServices
> 
> I had to install a bunch of nagios for 400000 polls. We ran some tests
> on
> bi-Xeon Dual-Core with 4Go. It seems the limit of Nagios 3 is around
> 25000
> services by server, but we didn't use the interface and the recording
> to
> database is with Perl scripts of our own. Taking 20000 services as a
> limit
> seem to be the right thing to do, perhaps less as 15000 with NDO.
> 
> Regards,
> -- 
> David Dumortier
> LINAGORA
> Service Management Monitoring
> 
> 
> 
> ------------------------------------------------------------------------------
> Return on Information:
> Google Enterprise Search pays you back
> Get the facts.
> http://p.sf.net/sfu/google-dev2dev
> _______________________________________________
> Nagios-users mailing list
> Nagios-users at lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/nagios-users
> ::: Please include Nagios version, plugin version (-v) and OS when
> reporting any issue. 
> ::: Messages without supporting info will risk being sent to /dev/null

I forgot to tell that we did run some stress tests in this customer with up to 30.000 services and Nagios supported it very well. We also use our own broker to store data in a MySQL database.

Regards,
Dario

------------------------------------------------------------------------------
Return on Information:
Google Enterprise Search pays you back
Get the facts.
http://p.sf.net/sfu/google-dev2dev
_______________________________________________
Nagios-users mailing list
Nagios-users at lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nagios-users
::: Please include Nagios version, plugin version (-v) and OS when reporting any issue. 
::: Messages without supporting info will risk being sent to /dev/null