[ntp:questions] Fwd: [ntpwg] RFC 5905 on Network Time Protocol Version 4: Protocol and Algorithms Specification

pc Paul.Croome at softwareag.com
Thu Jun 24 13:45:05 UTC 2010


At risk of stirring up a hornets' nest:

The RFC unequivocally states that "A primary server is
synchronized to a reference clock directly traceable to
UTC."

IMO, that is not a necessary condition. If I have a
hierarchy of NTP servers and clients with no external
connection to the Internet and I feed in Northern
Bongosooziland Spring Time (NBST) at the top of the
hierarchy, NTP will propagate that time throughout the
hierarchy. The only condition is that NBST must tick at a
rate of approximately 1 second per UTC second, otherwise the
finely-tuned FLL and PLL will not perform optimally.

Many users of this list have a requirement to synchronize a
number of machines within some user-defined limit, but they
don't care if they are all offset from UTC by a few minutes.
Time islands would seem to be a common use-case, and it's my
opinion that the RFC's assertion that genuine NTP networks
must be based on UTC is an unnecessary restriction. I
suggest that the RFC should mention that UTC-based NTP is
probably the most valuable use-case, and is the only form of
NTP that should be allowed on the Internet, whilst admitting
the existence of time islands.

Paul




More information about the questions mailing list