[ntp:questions] Poll timing

Danny Mayer mayer at ntp.isc.org
Sun Aug 19 03:53:28 UTC 2007

Jussi Kauppinen wrote:
> Thanks to everyone answering the topic this far!
> Paul.Croome at softwareag.com wrote:
>> Jussi,
>> If you have two requirements:
>>  1. You want to use NTP (or some derivative of NTP),
>>  2. You want tight control over the timing of the polling,
>> then I would suggest that these two requirements are mutually
>> incompatible.
>> Paul
> Paul,
> Why do you think there is incompatibility? If there is a function to 
> send a packet involving T1-timestamp to the server, why not it cannot be 
> sent at a specific moment? I mean that the daemon can wait until the 
> system time gets specific value (for example wait until the second unit 
> of the system time changes), and then send the packet.

This is just one process on the system. It has to get a timeslice to
notice that the timer has expired and then it has to finish whatever
else it is doing (like reading received packets and processing their
contents before it sees that it needs to send a packet. Naturally this
cannot happen instantaneously. I'm not sure why you are insistant on
this or why it's so important to have it sent at the exact moment. It
doesn't matter to NTP, so why does it matter to you? The packet is
timestamped when it goes out and not when the timer expires. Can you
explain what it is you are concerned about and why this is so important
to you? You don't get any better accuracy.


More information about the questions mailing list