Skip to content

Protocols: Roughtime, NTS pools, interleaved mode, NTPv5

ntpstats includes clients for the time protocols now being standardised or deployed, so they can be measured next to NTP and PTP. The weekly Live interop workflow runs every one of them against public servers from GitHub runners.

Protocol Status In ntpstats
Roughtime RFC 10049 (and the drafts and Google-Roughtime still deployed) ntpstats roughtime, ntpstats.roughtime
NTS pools draft-ietf-ntp-nts-keyexchange-pool-01 ntpstats query --nts --pool N, nts.pool_sessions
Interleaved mode RFC 9769 ntpstats query --interleaved, and detected in captures
NTPv5 draft-ietf-ntp-ntpv5-10 (and -09) ntpstats query --ntpv5, --probe-v5

Roughtime: signed time and provable misbehaviour

A Roughtime server signs its time (MIDP) and a radius (RADI) with an Ed25519 key. ntpstats checks every response:

  • the delegation certificate against the server's long-term key;
  • the response signature against the delegated key;
  • that MIDP lies within the delegation's validity (MINT ≤ MIDP ≤ MAXT);
  • the Merkle proof (PATH, INDX) that the server signed our request.

Servers are queried in sequence, twice, as the draft requires. Each nonce after the first is H(previous response ‖ random), so the chain of responses proves the order in which they were received. If an earlier response's earliest possible time is later than a later response's latest possible time (MIDP_i − RADI_i > MIDP_j + RADI_j), at least one server lied. The chain is then a malfeasance report that anyone can verify.

ntpstats roughtime                                  # the bundled server list
ntpstats roughtime roughtime.se time.txryan.com --json
ntpstats roughtime --list servers.json --report malfeasance.json --check-local
ntpstats roughtime --verify-report malfeasance.json # re-check a report's signatures and chaining

Each line shows the signed time, its radius, and the local clock's offset with its bound (radius + rtt/2). The intersection of all bounds is an authenticated interval for the local clock's error. It is coarse, about a second, because Roughtime timestamps are whole seconds. That is enough to catch a clock that NTP, PTP or GNSS has led astray: an NTP-disciplined clock outside that interval is wrong, whatever its own statistics say. --check-local exits with code 3 in that case.

The server list follows the JSON format of RFC 10049 section 8.3. The bundled list comes from the Roughtime ecosystem list. Check the keys against the operators' own publications before relying on them. Servers are added with host:port=BASE64KEY.

Compatibility: ntpstats offers versions 1 (RFC 10049), draft-08, draft-11 and draft-12+ (0x8000000c). On the wire, RFC 10049 differs from draft-19 in its version number and in the spelling of the two signature context strings (Roughtime v1 …, where the drafts have RoughTime v1 …); both spellings are accepted, the chosen version's own first, and the JSON output reports which one a server used (context). time.txryan.com moved to the RFC in October 2026, which ntpstats 3.8.0 and earlier report as a delegation signature that does not verify. Cloudflare's service documents draft-08 support. If a server does not answer the IETF request, ntpstats tries once with Google-Roughtime, the pre-IETF protocol: unframed messages, 64-byte nonces and hashes, microsecond timestamps. Some deployed servers still answer only that; in the live runs of October 2026, Cloudflare's does. The output says which protocol answered. Chained nonces and malfeasance reports work across both. Responses in the older-draft form are accepted too (nonce-only Merkle leaf, microsecond timestamps). The JSON output reports which form each server used (merkle_leaf).

Signature checks need cryptography: pip install 'ntpstats[nts]'.

NTS pools

With plain NTS, a client of a pool would share keys with the pool rather than with the time server. The NTS-KE pool draft instead has the pool's key-exchange server hand out one server per key exchange. A client that wants several independent sources sends NTP Server Deny records naming the servers it already has. ntpstats query --nts --pool 3 HOST performs three key exchanges this way and measures each assigned server with NTS:

ntpstats query nts-pool.example --nts --pool 3

The Trifecta Tech Foundation runs an experimental NTS pool at sectime.org, which gives its key-exchange host to users who accept its terms. Those terms allow personal use only and forbid preconfiguring the pool in software given to third parties, so ntpstats names no pool. Its earlier test host, srv.experimental.ntspooltest.org, has not resolved since September 2026. The live interop run probes a pool only when the repository variable INTEROP_NTS_POOL names one.

Interleaved mode (RFC 9769)

A server can only put an estimate of its transmit time in a packet before it has sent it. In the interleaved mode it sends the precise transmit time of its previous response in the next one, often a hardware or kernel timestamp. This removes the server's transmit-path error from the offset.

  • ntpstats query --interleaved HOST makes two exchanges. It reports whether the server answered interleaved and, if it did, the offset computed with the precise timestamp.
  • Captures: interleaved responses (origin equal to the request's receive timestamp) are recognised in pcap/pcapng files. Each exchange is recomputed with the precise transmit time from the following response and marked in the interleaved column. This works for chrony clients using xleave.

NTPv5

The NTPv5 client follows draft-ietf-ntp-ntpv5-10 (29 October 2026) and falls back to draft-09. Draft-10 removes the leap indicator and the leap-smeared UTC timescale, turns flag 0x1 from synchronized into unsynchronized, shortens reference IDs to 56 bits and adds Time Sources extension fields. A server must ignore a draft name it does not know, so the client asks in draft-10 first, retries in draft-09 when there is no answer, and reads the flags by the draft that answered (draft in the result). It covers timescales (UTC, TAI, UT1, and leap-smeared UTC in draft-09), eras, cookies and the NTPv4 upgrade probe. See CLI and the Live interop results for which public servers answer.