ntpstats.research¶
Research and time-laboratory data: CGGTTS, RINEX clock, BIPM Circular T, RIPE Atlas, NTP Pool.
- CGGTTS (BIPM/CCTF generic GNSS time-transfer format, V2E): one line per
satellite track. REFSYS (reference − GNSS system time, 0.1 ns) is averaged
per epoch into a series; line checksums are verified. :func:
common_viewdifferences two stations, satellite by satellite (common view) or epoch by epoch (all in view), giving UTC(A) − UTC(B). - RINEX clock (IGS
.clk, 3.0x): one series per receiver (AR) or satellite (AS) clock bias, in seconds against the file's time scale. - BIPM Circular T, section 1:
UTC − UTC(k)per laboratory every five days. Concatenate several issues into one file for a long series. - RIPE Atlas NTP results (JSON from the public API): one series per probe and server. Atlas reports offsets as local − server; they are negated to the ntpstats convention (server − local).
- NTP Pool score logs (
/scores/<ip>/log?monitor=*CSV): one series per monitor, with round-trip time and score.
cggtts_tracks(lines)
¶
Header fields and one record per track (dict of columns) of a CGGTTS file.
Several files concatenated together (as in the BIPM archives) are accepted.
parse_cggtts(lines, name='cggtts')
¶
REFSYS (reference − GNSS time) averaged over the satellites of each epoch.
common_view(a_lines, b_lines, mode='cv', min_sats=1, name=None)
¶
REF(A) − REF(B) from two CGGTTS files.
mode="cv": common view, REFSV differences of the same satellite at the
same epoch, averaged per epoch (satellite clock errors cancel).
mode="aiv": all in view, difference of the per-epoch REFSYS means
(needs no common satellites; relies on precise orbits and clocks).
parse_rinex_clock(lines, name='rinex-clock', types=('AR', 'AS'), clocks=None)
¶
Clock bias series from a RINEX clock file (one per receiver/satellite).
parse_circular_t(lines, name='circular-t', labs=None)
¶
UTC − UTC(k) per laboratory from section 1 of one or more concatenated Circular T issues.
parse_ripe_atlas(text, name='atlas')
¶
NTP measurement results from the RIPE Atlas API (JSON list or JSON lines).
group_summary(series, by='peer')
¶
Offset distribution per group (peer: server, probe: vantage point) across many series.
parse_ntppool(lines, name='ntppool')
¶
NTP Pool monitoring log CSV: one series per monitor (offset in s, rtt in ms).
parse_interop(lines, name='interop')
¶
The ntpstats Live-interop dataset (JSON lines, data/interop/): one series per test and server.
Failed probes carry no offset; they are counted in meta (probes,
failures, availability) and their times in meta["failure_times"].
interop_summary(lines)
¶
Per test and server: runs, availability, median offset/delay, last error (from the raw records).
interop_trend(lines, shift=0.001, consensus_tests=CONSENSUS_TESTS)
¶
Per test and server, how its offset, delay and availability evolve over the runs of the dataset.
The runner's clock is not a reference and changes between runs, so each
offset is taken relative to the run's consensus: the median offset of
all successful NTP-family probes (consensus_tests) in that run. The
runner's own clock error cancels; what remains is the server's offset from
the other public servers plus the path asymmetry of that run.
Each row has the runs where the server was probed, its availability
overall and in the last run, the relative offset (median, first, last and,
with three runs or more, the least-squares slope per week), the delay at
the first and last run, and status:
gone
probed successfully before, failing in the last run;
new
first probed after the dataset's first run;
shift
the last relative offset differs from the median of the earlier ones
by more than the two runs' error bounds plus shift seconds
(default 1 ms). The bound of an NTP exchange is half its round-trip
delay (the offset error of a correct server and client cannot exceed
it, whatever the path asymmetry); of a Roughtime response, its radius.
A path change alone cannot produce a shift; a server whose clock moved
can;
ok
none of these.
timeline lists, per run, [run, ok, relative offset, delay, bound].