[ntp:questions] Interesting results with PPS

Mauro Fiacco Mauro.Fiacco at ipaccess.com
Wed Feb 7 15:51:14 UTC 2007


Hello to All,

I have some interesting results using PPS to share.

Note that:
- all my boxes run FreeBSD (6.1 and 6.2).
- NTP v 4.2.2 
- PPS is generated by a very stable Rubidium source
- NTP server #1 provides the timebase (at the moment I cannot receive
  GPS in the building). All the others synchronise to it.
- NTP server #1 provide a stability better than 10ppb
- All servers use PPS kernel discipline (not ntpd PPS driver!)
- The polling period was increased to 14 in order to maximise stability
  results (...other tinker parameters were also modified...)

Setup n. 1:
-----------

- NTP server #2 employs the same PPS signal as NTP server #1
- Time is synchronised by ntpd, only NTP server #1 is configured
- The two servers sits on the same Ethernet switch
- results of ntpq -crv:

status=0644 leap_none, sync_ntp, 4 events, event_peer/strat_chg,
version="ntpd 4.2.2p3 at 1.1577-o Wed Sep 27 20:58:11 UTC 2006 (2)",
processor="i386", system="FreeBSD/6.1-RELEASE", leap=00, stratum=2,
precision=-18, rootdelay=0.279, rootdispersion=5.003, peer=46944,
refid=xxx.xxx.xxx.xxx,
reftime=c97455bc.7f8c3ece  Wed, Feb  7 2007 13:42:20.498, poll=6,
clock=c974569e.33770463  Wed, Feb  7 2007 13:46:06.201, state=4,
offset=0.949, frequency=-42.494, jitter=0.024, noise=0.028,
stability=0.018, tai=0

NTP server #2 provides excellent accuracy and stability, to be expected
given the low jitter. 

Setup n. 2:
-----------

- NTP server #3 employs an unsynchronised version of the PPS signal
  available to NTP server #1
- Time is synchronised by ntpd, NTP server #1 and #2 are configured
- The NTP server #3 connects to the other servers via a different ISP
  provider (albeit these are business grade connections)
- Note that the peer filter algorithm always selects NTP server #1
  (which is the stratum 1 after all)
- results of ntpq -crv:

status=0644 leap_none, sync_ntp, 4 events, event_peer/strat_chg,
version="ntpd 4.2.2p3 at 1.1577-o Wed Jan 31 10:15:55 UTC 2007 (5)",
processor="i386", system="FreeBSD/6.2-RELEASE", leap=00, stratum=2,
precision=-19, rootdelay=49.455, rootdispersion=7.286, peer=47608,
refid=xxx.xxx.xxx.xxx,
reftime=c97427ab.9d96ca3f  Wed, Feb  7 2007 10:25:47.615, poll=14,
clock=c9745908.ea5119a3  Wed, Feb  7 2007 13:56:24.915, state=4,
offset=1.459, frequency=23.129, jitter=5.380, noise=2.723,
stability=0.057, tai=0

The NTP server #3 provides good offset, stability is affected by higher
network jitter.

Setup n. 3:
------------
- NTP server #4 employs the same unsynchronised version of the PPS
  signal as NTP server #3
- Time is synchronised by ntpd, only NTP server #1 is configured
- The NTP server #4 connects to NTP server #1 via the same network path
  as NTP server #3 
- results of ntpq -crv:

status=0644 leap_none, sync_ntp, 4 events, event_peer/strat_chg,
version="ntpd 4.2.2p3 at 1.1577-o Wed Jan 31 08:42:18 UTC 2007 (3)",
processor="i386", system="FreeBSD/6.2-RELEASE", leap=00, stratum=2,
precision=-19, rootdelay=41.177, rootdispersion=17.558, peer=27086,
refid=xxx.xxx.xxx.xxx,
reftime=c9741aed.6bcdf009  Wed, Feb  7 2007  9:31:25.421, poll=14,
clock=c9744ef2.c5657e88  Wed, Feb  7 2007 13:13:22.771, state=4,
offset=-11.474, frequency=27.642, jitter=5.628, noise=3.958,
stability=0.129, tai=0

This experiment is very interesting... Although the network path is the
same as NTP server #3 (measured jitter is the same), the stability of
the servers is constantly worse (twice as big!). 

The only difference is that NTP server #3 employs two servers for
time synchronisation, while NTP server #4 employs only one server. 

I think the only difference is the working of the clustering algorithm.
The use of two NTP servers improves stability and accuracy when compared
to only one servers.

Regards,

Mauro

-- 
Mauro Fiacco 

ip.access Ltd                      
URL: www.ipaccess.com  




More information about the questions mailing list