Here are some numbers from the US pool. These are not long
term stats just quick and dirty numbers obtained by loading the
entire US pool into ntpd with "minpoll 10 noselect".
The test was run from a stratum one server that is less that
15us offset from it's GPS/PPS reference and less that 250us
offset from three other stratum one servers.
After loading the config ntpd was allowed to run until all
operational hosts had 377 status. The output of ntpq -np
was then loaded into a spreadsheet and the following stats
are the result.
All N=169 Delay Offset Jitter
Median 81.259 0.245 1.378
Standard Dev 48.361 37.497 101.965
Stratum 1 N=19
Median 86.75 -3.074 1.829
Standard Dev 102.876 85.101 297.616
Stratum 2 N=101
Median 84.014 0.352 1.231
Standard Dev 36.462 19.124 25.914
Stratum 3 & 4 N=49 (S3=45, S4=4)
Median 67.699 0.867 1.518
Standard Dev 34.744 27.018 14.756
Interestingly the pool as a whole was very close to
the test machine. The worst servers in the pool in
terms of offset all claimed to be stratum 1. Offsets
between +16ms and -283ms were observed among the stratum
one servers. Six "stratum one" servers were outside
+/- 10ms. Of these two appear to be running openntpd,
one is synced to WV15 as it's only time source, one is
claims a truetime refclock as it's only time source,
one uses CDMA and one uses GPS. The remaining Stratum
one servers mostly use GPS with a minority using CDMA and
one WWVB.
The stratum two, three and four servers as a group were
much better showing less jitter and a good overall
stability. The outliers in terms of offset correlated
strongly with high jitter numbers suggesting that the
network is the main cause of inaccuracy in S2/3/4 pool
members as opposed o
More analysis to follow.
John
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
