[ntp:questions] NTP Newbie, can anyone help out ?

Richard B. Gilbert rgilbert88 at comcast.net
Thu Nov 30 22:41:39 UTC 2006


apexi.200sx at hotmail.co.uk wrote:

> Hi All,
> 
> I am currently at a site where we have workstations running an
> application program that requires accurate time.  The application
> writes a date time stamp to a database, the dts is generated locally on
> the machine. 
<snip>

> All the client workstations currently use xntpd for time keeping, and
> we have a time server etc. all already configured, thing is we seem to
> still be getting issues where clients aren't having their time adjusted
> correctly, or maybe quickly enough ?

Details please??  You can't fix it until you know what's happening!!
Include the version of ntpd you are running.  The name "xntpd" was 
changed to "ntpd" several years ago which suggests that you may be 
running a very old version.

<snip>
> We have hundreds of workstations and sometimes they may have been
> switched off for a peroid of time, or the CMOS battery may have died -
> does this affect NTP on the whole ?

If you start ntpd in such a way as to ensure that the clock is correct 
or very close to correct initially it should not make much difference.
With modern versions you use a -g switch to have it set the clock as it 
starts.  With older versions you need to use ntpdate to set the clock 
before starting ntpd.

Note that cold starting ntpd can require as much as thirty minutes to 
achieve tight synchronization.  The frequency of the crystal oscillator 
that drives the clock is dependent on the temperature which tends to 
increase rapidly when the machine is started.  It's best to leave the 
machines running if possible.  If the electric bill is a problem, try to 
have someone turn on the machines thirty minutes before the work day starts.

> 
> Basically, could someone give me a brief primer on NTP from a
> protocol/algorithm/operational point of view (if possible)
> 
There are links to a lot of documentation on the NTP web page:
http://www.ntp.org/

> I think we could be better with just a very basic SNTP ?, or just
> UDP/TIME implementation

SNTP is pretty crude!  The times on two workstations could differ by as 
much as two seconds!  With ntpd the difference between any two 
workstations should be less than one second; a lot less!  With a 
properly designed and operated NTP subnet, it's possible keep a whole 
herd of machines within 10 to 20 milliseconds or better.
> 
<snip>




More information about the questions mailing list