Windows – Ntpd excessively inaccurate

The NTP algorithm includes information to allow you to calculate and fix the drift in your server's clock. NTPD includes the ability to use this to keep your clock in sync and will run more accurately than a clock on a computer not running NTPD. NTPD will also use several servers to improve accuracy.

ntpdate does not keep any state to perform this service for you so will not provide the same kind of accuracy. It will allow you to provide it with a list of servers which it will use to attempt to provide you with a better result but this is no substitute for the sophisticated algorithms provided in NTPD that track your drift from each of the servers over time.

NTPDATE corrects the system time instantaneously, which can cause problems with some software (e.g. destroying a session which now appears old). NTPD intentionally corrects the system time slowly, avoiding that problem. You can add the -g switch when starting NTPD to allow NTPD to make the first time update a big one which is more or less equivalent to running ntpdate once before starting NTPD, which at one time was recommended practice.

As for security concerns, ntp servers do not connect back on uninitiated connections, which means your firewall should be able to tell that you initiated the ntp request and allow return traffic. There should be no need to leave ports open for arbitrary connections in order to get NTPD to work.

From the ntpdate(8) man page:

ntpdate can be run manually as necessary to set the host clock, or it can be run from the host startup script to set the clock at boot time. This is useful in some cases to set the clock initially before starting the NTP daemon ntpd. It is also possible to run ntpdate from a cron script. However, it is important to note that ntpdate with contrived cron scripts is no substitute for the NTP daemon, which uses sophisticated algorithms to maximize accuracy and reliability while minimizing resource use. Finally, since ntpdate does not discipline the host clock frequency as does ntpd, the accuracy using ntpdate is limited.

The value that ntpstat displays after "time correct to within" is the root dispersion + root delay / 2. ntpq -p doesn't show the "root dispersion" run ntpq -c rl instead.

Notwithstanding, it is clear that the main source of the lack of accuracy is the dispersion rather than the delay (which is only 0.964).

The dispersion is the "nominal error relative to the primary reference source." I've briefly looked through the NTPv4 RFC and this is what it has to to say:

The dispersion (epsilon) represents the maximum error inherent in the measurement. It increases at a rate equal to the maximum disciplined system clock frequency tolerance (PHI), typically 15 PPM. 1 PPM is equal to 10^(-6) seconds/second.

To use rrdtool terminology dispersion isn't a gauge but rather a counter. Seeing a large value may not indicate anything is wrong.

Alas, I wasn't able to understand the ntp algorithm well enough to see how to make this number smaller. I've noticed this value is reset occasionally. I don't know why.