[ntp:questions] Local time sources

Richard B. Gilbert rgilbert88 at comcast.net
Mon May 10 01:33:46 UTC 2010

Jaap Winius wrote:
> Hi all,
> When configuring a typical ntpd system, the recommended minimum number of 
> server entries in ntp.conf for NTP time sources seems to be three, 
> apparently because this way ntpd can calculate an average time, eliminate 
> the server furthest from the average, and then still have two left for a 
> final selection that depends on other values, such as jitter.

Not three but four!  If you have only three servers configured and one 
fails you have the worst possible configuration.  Is it not written that 
a man with two clocks can never be certain what time it is?

> However, what is the recommended minimum number for systems that only 
> work with local time sources, such as GPS devices and time signal radio 
> receivers, that each supply their own timing information as well? Is one 
> always enough, or are there still advantages other than redundancy in 
> having more than one?

Four, five, and seven are the magic numbers to protect against: one, 
two, and three failed servers, respectively.  A "failed server" is one 
that is either responding with an incorrect time or not responding at all.

If you have legal requirements to meet with respect to time stamps you 
should have four, five, or seven servers, each with an accurate and 
reliable source of time.

GPS timing receivers are available for $100 or less (Garmin GPS-18LVC) 
and are accurate to within 50 or 100 nanoseconds.  You lose some of that 
accuracy while getting the time into a computer.  Just about any PC can 
be used as a time server

There are other sources of time: VLF (WWVB) and HF radio (WWV, WWVH) 
broadcast by the National Institute of Standards and Technology (NIST).
Canada's CBNRC also has has an HF radio broadcast (CHU).  If you have 
money to burn you can have an atomic clock of your very own!

More information about the questions mailing list