[EAS] EAS time stamp usage

[Sean Donelan]

On Mon, 19 Dec 2011, Jim McKinnon wrote:
> 2.) On another broadcast list, there was a discussion about the NIST time
> servers sending out time signals which were inaccurate by about ten minutes
> or so for a short period of time.

Judah Levine of NIST posted an explaination of what happened at the NIST
Internet time servers.  One of those double hardware failures in 
redundant systems that should not happen at the same time happened. 
Although it may look easy from the outside, maintaining clock standards 
is a bit complicated.  If you have to replace an EAS box with a cold spare 
after a catastrophe, synchronizing clocks may be a bit of a challenge.

Which brings up another point with NTP and the new CAP boxes.

Vendors shipping equipment using a few popular time servers as defaults in 
their configurations.  Even without double hardware failures, this leads 
to overloading of those popular time servers, especially around the time 
of leap seconds. NTP has flags and checks to detect bad clocks under 
those conditions, when properly configured. But very few NTP 
configurations are done properly though.

Vendors should not ship default configurations using the names (or IP 
addresses) of time servers the vendor doesn't operate or should follow the 
so-called NTP rules of engagement for default configuration settings.

The NTP folks have a FAQ about the issue of vendor default configurations 
for NTP servers.  And one possible way vendors can ship default 
configurations.  You may not like the pool.org approach, for example 
Microsoft choose to use time.windows.com as its default instead.

http://www.pool.ntp.org/en/vendors.html