[time-nuts] can of worms: time-of-day in a community radio station

[Hal Murray]

Just out of curiosity, how many PSc/laptops/servers does it take to run a 
radio station?


eric at scace.org said:
> What NTP software should be used on Windows OS machines? Linux servers?

There is a version of ntpd available for Windows.  Meinberg builds and 
distributes it.

Most Linux distros come setup to use ntpd with servers from the pool.

It's impolite to use the pool (or other servers) if you have more than a 
handful of clients.  Setup a pair (or more) of servers to use the pool and 
point your internal clients at them.

You may be able to configure your dhcp server to tell clients which NTP 
servers to use.



> desktops and servers: better NTP configuration NTP checks: at least hourly? 

I treat things like "at least hourly" as an indication of suspicious software.

There are a lot of packages that poll a server and smash the local clock.  
They usually have an option to set how often to do that.

The real key to decent timekeeping on a computer is to calibrate the crystal 
offset.  Modern OSes have a way to set that knob.  ntpd calls it drift.  It's 
saved in a file so it can be reloaded on reboot.  (Units are PPM.)  ntpd 
adjusts the polling interval automagically looking for the bottom of the adev 
graph with a cap at 1024 seconds.

The crystal is a thermometer.  On modern systems, the temperature depends on 
the workload as well as the air conditioning.


>   * some systems are on in-house Ethernet; others on in-house wi-fi

How good is your WiFi?  (mine is crappy)

The other consideration in this area is how good is your Internet connection?  
If it is saturated by a big download or upload, that may confuse ntpd.  ntpd 
assumes the network delays are symmetric.  It also filters out quirky samples. 
 The filter uses the last 8 samples so you need a long-enough download/upload 
to cause problems.  (There is a huffpuff option to try harder.)


>      − no smearing (minimum requirement)
  Beware of Google's NTP servers.

>      − accurate leap second display (desirable but unlikely to be achievable)

What would anybody do if they saw a blinking "LEAP" on a display?

The NTP protocol distributes a leap-tonight indication.  It's fairly easy to 
query your ntpd server to see if that will happen.

NIST and others distribute a file with a list of leap seconds, including the 
next one if it has been announced.  (ntpd can use that file to introduce local 
leaps and/or override leaps from the net.)


>          … use time.nist.gov <http://time.nist.gov/> and similar
> higher-quality NTP servers 

They are high quality in the legal sense, but the network between you and the 
server may add a lot of noise.

You may be able to find good local servers.  Ask around.  What do the local 
geeks use?  (There is probably an overlap between radio geeks and network 
geeks.)

-----------

How often does your internet link go down?

-----------

The package that provides ntpd probably includes ntpq.  You can use it see 
which servers ntpd is using, the round drip time to each server, and the 
offset to each server.  Try the peers command or ntpq -p <server> (defaults to 
localhost).

ntpd is a server as well as a client.  You can setup one system to monitor all 
the other systems you are interested in.  In your config file, add a server 
line with noselect.  That way, ntpq -p on one system will give you a quick 
health indication of the others.  You can also collect make graphs and/or will 
have logs if you want to investigate glitches.

-----------

You can setup a Raspberry Pi with a GPS as a time server.  That will give you 
a good place to stand if you want to see how your NTP servers are doing.  (You 
need a GPS Hat.  The low cost GPS mice that connect via USB don't provide PPS 
and even if they did, USB adds a lot of noise.)

It may not work with the antenna indoors.  Maybe near a window?  Or for a 
temporary hack, use a long extension cord and long ethernet cable and put it 
outside.

-- 
These are my opinions.  I hate spam.