NTP is a very accurate and non-invasive program. Edit /etc/rc.config.d/netdameons, and set the lines:
export NTPDATE_SERVER=
XNTPD=0
export XNTPD_ARGS=
Change the first line to point to the IP-addr or hostname of your primary NTP server. Change the second line from =0 to =1 to enable xntpd on reboot. Now, every reboot will sync exactly to the NTP server. A reboot is the fastest and safest way to reset the clock. For example:
export NTPDATE_SERVER=myntpsource.mydomain.com
XNTPD=1
export XNTPD_ARGS=
However, you don't want time changing visibly on ANY server, especially a database server, so the NTP daemons will actually change the CPU clock speed (more accrately, it changes the clock divisor very slightly). This means that a 1 minute mismatch in time can take up to an hour to sync.
If you have the latest patch for NTP, there is a new option for ntpdate (-B) which will slew the clock slowly, essentially what xntpd does. Use ntpdate -B when the clock is seriously out of sync, like 20 minutes or more.
NTP is an amazingly complex and accurate protocol. It starts out with a 64 byte message each way every 64 seconds and adjusts the clock just a fraction of a second. This means that not a single 1-second tick will be missed. This keeps database programs vary happy.
Once the clock is within 128 milliseconds, the protocol backs off to once every 128 seconds, then 256 and so on, out to a maximum of about 17 minutes per check. As long as the local clock is within 128 milliseconds, it won't poll any faster.
This is why a very small computer can act as a timeserver for THOUSANDS of clients. As a rule, sysadmin must NEVER change the time of day on a Unix box that is running databases or other time-sensitive programs. You might even put a wrapper around the date command to reject any attempt to change the time (but leave the other functions alone).
Bill Hassell, sysadmin
