[ntp:questions] Re: Question about ntp log messages and logging interval
rmoore at rmoore.dyndns.org
Wed Sep 13 00:12:12 UTC 2006
On Tue, 12 Sep 2006, David Woolley wrote:
> In article <R4adnZdGoqCNZpvYnZ2dnUVZ_q2dnZ2d at comcast.com>,
> Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
>> Jeff Boyce wrote:
>>> 1. I am wondering if this is an indication of a properly (or improperly)
>>> running ntp system?
> Time steps in the same direction are an indication of a failing system.
> Either, your crystal is out by more than 500ppm, or, more likely,
> especially for Red Hat, is that you are losing clock interrupts. A
> contributory factor tends to be the use of HZ=1000 and IDE device
> drivers with interrupt latencies of more than 1ms. It is strongly
> reccommended that HZ be no more than 100.
I can expand on this comment. There are a couple ways to check what your
HZ value is set to if you don't know it.
1. The easy way: check /proc/config.gz
If you have the appropriate support in the kernel, this file will show
what options were used to compile the kernel. Check for HZ. Here's an
example on my Linux system:
~> gzip -dc /proc/config.gz | grep HZ
# CONFIG_HZ_100 is not set
# CONFIG_HZ_1000 is not set
So my HZ is 250.
2. If you don't have that, check your timer interrupts.
This should always work, check the timing on your timer interrupts.
~> grep timer /proc/interrupts ; sleep 30; grep timer
0: 146358116 IO-APIC-edge timer
0: 146365617 IO-APIC-edge timer
Both methods come up with 250.
Finally, I thought there was an issue where when ntp is compiled on Linux,
up the value of HZ in the kernel headers. So if ntp was compiled on a
machine with a different value of HZ than the machine you run it on, it
won't work right. I don't know this for a fact, it's something I thought
I heard before.
More information about the questions