ntp config tech note

Fri May 21 04:16:41 UTC 2004

Robert E. Seastrom wrote:
> 
> "Hannigan, Martin" <hannigan at verisign.com> writes:
> 
> 
>>That's NTPv4 isn't it? 
>>
>>I also prefer to use three peers vs. two. Always an odd number,
>>greater than 1. Assumptions can't be made about the mathematics 
>>behind time, but in a reference model, odd numbers are better.
> 
> 
> Actually, three is not enough; Mills says at least four.  Diversity in
> manufacturer (and controlling organization if you can spare the
> cycles) is a big big plus. You may wish to read Dr. Mills' post to
> comp.protocols.time.ntp in the wake of the TrueTime bug of the
> 2001->2002 new year:
> 
> http://groups.google.com/groups?hl=en&selm=3C32924F.994E1D01%40udel.edu

If you're really paranoid, diversity in reference sources should also be 
considered.  You should have more than one stratum-1, and as a group 
they should get time from more than one of [GPS, 
WWV/WWVB/DCF77/CHU/JJY/ETC., USNO, ACTS, etc.] and your stratum>1s 
should get time from multiple stratum-1s of similarly diverse references.

Many NTP folk look down their nose at the radio sources, since GPS is 
more accurate.  But if you already have a GPS stratum-1, then perhaps 
your next stratum-1 should be WWVB and friends, or you should have a 
backup assocation with someone who does.  And remember that CDMA gets 
its time from GPS, so it doesn't "count" as a diverse source.  Like I 
said, if you're really paranoid...

michael