ntp performance - NTP
This is a discussion on ntp performance - NTP ; hello,
I'm running a ntp stratum 1 server, ntp.i2t.ehu.es, and i'm having
some variations with the offset and frequency. These variations
depends on the day and climate, for example when it's very foggy. But
every day at the same time, ...
-
ntp performance
hello,
I'm running a ntp stratum 1 server, ntp.i2t.ehu.es, and i'm having
some variations with the offset and frequency. These variations
depends on the day and climate, for example when it's very foggy. But
every day at the same time, between 0 and 10000 UTC seconds normally
at 3.00 am local time, there is a great peak of the offset. Oddly at
this time FreeBSD runs daily periodic tasks using cron and periodic
applications. My question is if a high CPU usage could produce a
variation of the offset in a FreeBSD machine. Did anybody have this
trouble?
The graphic could be seen at http://services.i2t.ehu.es/ntp
Thanks,
--
Christian Pinedo Zamalloa
-
Re: ntp performance
wrote in message
news:1181637439.758438.131310@j4g2000prf.googlegro ups.com...
[...]
> every day at the same time, between 0 and 10000 UTC seconds normally
> at 3.00 am local time, there is a great peak of the offset. Oddly at
> this time FreeBSD runs daily periodic tasks using cron and periodic
> applications. My question is if a high CPU usage could produce a
> variation of the offset in a FreeBSD machine.
Not actually CPU usage but disk usage. Are they configured for DMA?
PIO mode can notoriously result in lost (timer) interrupts during
peak activity.
Groetjes,
Maarten Wiltink
-
Re: ntp performance
christian.pinedo.zamalloa@gmail.com writes:
> My question is if a high CPU usage could produce a variation of the
> offset in a FreeBSD machine. Did anybody have this trouble?
One of the first things I tend to do when setting up a machine is slap
a "nice -20" in front of any non-interactive, long-running programs in
the startup files. Good candidates are cron, sendmail/postfix,
thttpd/httpd, innd etc. The same thing is done for make(1) via an
alias. My feeling is that there is no good reason why these programs
should get first dibs on CPU cycles. They will still get all the
cycles they want, but they might be delayed a few milliseconds to get
them. Maybe this trick will help you? I never noticed it if helped
ntpd, but it sure helped interactive typing response many years back
when I started doing this. (Although the CPU's were 100x slower then,
so the effect was much more noticeable.)
-wolfgang
--
Wolfgang S. Rupprecht http://www.wsrcc.com/wolfgang/
IPv6 on Fedora 7 http://www.wsrcc.com/wolfgang/fedora/ipv6-tunnel.html
-
Re: ntp performance
You might want to look at the -N flag for ntpd.
H
-
Re: ntp performance
On 2007-06-13, Harlan Stenn wrote:
> You might want to look at the -N flag for ntpd.
>
Finally I have modified the ntpd startup script of FreeBSD to start the
service with a priority of -15. I couldn't use the flag "-N" because it
seems that the version of ntp installed in FreeBSD 6.1 doesn't support
it.
I will see in the following days if we get a more precise clock. Thanks
for all your help and comments,
--
Christian Pinedo Zamalloa
-
Re: ntp performance
>>> In article , Christian Pinedo Zamalloa writes:
Christian> On 2007-06-13, Harlan Stenn wrote:
>> You might want to look at the -N flag for ntpd.
Christian> I couldn't use the flag "-N" because it seems that the version of
Christian> ntp installed in FreeBSD 6.1 doesn't support it.
It might be '-N high' in that version.
H