NTP Question to the NTP Gurus

[Kenneth_18]

Hi,

We have our server configured as an ntp server and getting the clock from a gps source via the serial port.

When we get a ntp response from the server in our ntp client, in the ntp response packet, the "root dispersion" value is somewhere greater than 300 ms.

From my understanding, root dispersion means the maximum error the local clock has deviated from the referrence clock. But as since this ntp response came from the ntp server, does this mean that this value is the maximum error that the ntp server has occured relative to the GPS Clock?

If I do an ntpq then rl command in the ntp server, the root dispersion value is also in the 300ms range.

I was just wondering where this 300ms root dispersion value came from as the local clock of the ntp server is directly synchronized to the gps clock. Could it be that this 300ms root dispersion value come the gps and it passes it down to the ntp server and then is passed down further to the ntp clients?

How can I have this root dispersion value decreased to less than 100ms?

Thanks in advance for any insights.

[Bharat Katkar]

HI kenneth,
See if the link below helps.

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=550229

Regards,

[Steven E. Protter]

Shalom Kenneth,

ntp won't adust time on a client if the clients time is of too far. I think you can try and modify the variable and restart xntpd.

SEP

[Bill Hassell]

NTP is an ultra-precise time sync method which is always attempting to sync the local system to less than 128ms of the reference clock. Most GPS clocks are not nearly as stable as Internet time sources and using the serial port means 100's of ms just to transmit the sync request and then receive the response. So reducing the dispersion level will require a high speed connection to a highly stable clock source. Now NTP will not be jumping your clock all over the place...NTP will see that exactly 86400 seconds occur every day. All the systems that sync to your NTP server (over the local LAN) will be accurate to less than 128ms to the main server. Since they are all in sync, there is no problem with a higher dispersion unless you compare the time to a super-accurate time source. We're talking about less than one second of uncertainty which is certainly more accurate than the internal clocks of most servers. Check out the details at: www.ntp.org

[Kenneth_18]

Not the answers I was looking for but definitely pointed me to some other resource I might find the answer to.

Thanks again guys!