2007/9/17, Dennis Hilberg, Jr. <[EMAIL PROTECTED]>:
>
> Thierry wrote:
> >
> >
> > 2007/9/15, Chuck <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>>:
> >
> >     On Saturday 15 September 2007, Chip Cuccio wrote:
> >      >
> >      > On Sep 15, 2007, at 11:45 AM, Chuck wrote:
> >      >
> >      > > what are most of you using to monitor local ntp servers?
> >      >
> >      >
> >      > A combination of Wayne S.'s scripts (
> http://www.schlitt.net/scripts/
> >      > ntp/index.html), Dennis H.'s RRDtool scripts (http://
> >      > saturn.dennishilberg.com/ <http://saturn.dennishilberg.com/> ),
> >     and my own scripts/hacks.
> >      >
> >      > The scripts by Wayne and Dennis' are excellent for my monitoring
> >      > purposes.
> >      >
> >      > --
> >      > Chip Cuccio
> >      >
> >     excellent. thanks!! looks like i have a bit of 'playtime' ahead :)
> >      >
> >      > _______________________________________________
> >
> >
> >
> >
> > I still have issues with those scripts regarding clients stats. I think
> > tcpdump is a bit CPU intensive from time to time.
> >
> > I'm currently using iptables/ULOG and a little script to generate this
> > kind of stats.
>
> I don't have any problems with tcpdump and CPU load as long as the client
> count stays relatively low.  But when I get those 60k+ active client peaks
> the system does get loaded down.  I don't generate load statistics, but on
> a
> couple occasions I've been logged into the system during a huge peak and
> the
> load average was in the 2.25-2.5 range.
>
> For comparison, a typical active client count for my ntp server is in the
> 5000-5500 range, and ntp_clients and tcpdump show 0.0 %CPU most of the
> time
> with a 0.00 load average.
>
> Since this new DNS system has been unveiled for the pool, hopefully those
> large peaks are a thing of the past.
>
> --
> Dennis Hilberg, Jr. <[EMAIL PROTECTED]>
> NTP Server Information: http://saturn.dennishilberg.com/ntp.php
> _______________________________________________
>
>
I agree it's only when there are spikes. Regarding spikes issue  that should
be better with the new DNS system.

I'm still trying to avoid the tcpdump solution for these reasons :

 -  promiscuous mode is not safe
 -  tcpdump should be use only for troubleshooting
 -  tcpdump is capturing the whole packet where we only need a part of it.

I looked a bit around and the only proper solution i found was to use
iptables and ULOG. My concern now is that's generating io.

If you have any cool idea but patch ntpd to have (active|inactive|abuse)
clients numbers, pls reply !

--Thierry
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to