NTP sync different

[R.SRIDHAR]

Hi
what is the time deviation can servers have when all servers are being synced with NTP server

[James R. Ferguson]

Hi:

The lastest NTP version can maintain time to within about 10-milliseconds using Internet sources.

Regards!

...JRF...

[R.SRIDHAR]

when i checked in the servers we are having this different's. Is this okay or normally how much deviation will be there between the NTP server and the client.

Sun Jul 19 18:25:28 MDT 2009

Sun Jul 19 18:25:29 MDT 2009

Sun Jul 19 18:25:30 MDT 2009

Sun Jul 19 18:25:30 MDT 2009

Sun Jul 19 18:25:32 MDT 2009

Sun Jul 19 18:25:33 MDT 2009

[Bill Hassell]

The actual accuracy is very dependent on the dispersion between the NTP source and the clients. The protocol is capable of extreme accuracy with a very small message stream. When the network is stable, 1/10th of a second is normal. Using several (notice: several) sources on the Internet such as the NTP pool servers, accuracy less than one second is also fairly normal, but it all depends on the noise (jitter, packet loss, etc) along your connection path.

Here is a reference: http://en.wikipedia.org/wiki/Network_Time_Protocol

[rick jones]

Your checks look like a series of remsh'd date commands, which means you will have been at the mercy of the network and DNS etc to get those running.

Instead I would suggest pointing an ntpq command at each system (or running one there) and looking at the peers output to see what ntp reports as its difference from the time server to which it is synced.

[Bill Hassell]

> Sun Jul 19 18:25:28 MDT 2009
> Sun Jul 19 18:25:29 MDT 2009
> Sun Jul 19 18:25:30 MDT 2009
> Sun Jul 19 18:25:30 MDT 2009
> Sun Jul 19 18:25:32 MDT 2009
> Sun Jul 19 18:25:33 MDT 2009

This is a very bad test for sync as Rick says. I use a similar script to query a lot of servers. It takes almost 1 second to get the date back from each system -- probably the same for your test. Therefore the servers appear to be in excellent sync -- your date command is far too slow across the network. NTP is accurate to a small fraction of the query you are running.

As Rick said, you want to run /usr/sbin/ntpq -p to query each server. The easiest numbers are poll and reach. Poll is the time between sync requests in seconds. 64 is the worst which means that NTP has to keep trying to get a stable sync. reach should always be 377 -- anything less means the NTP server is erratic or occasionally not reachable.