NTP Sync Quality — NTP Precision Accounting Test

Test Matrix

Summary

• All hosts are c7i.metal-48xl in eu-west-1a, Fedora 43 Cloud, running kernel 7.1.0-rc4+
• NTP via chrony (default Fedora 43 configuration, no tuning) to Amazon Time Sync (stratum 2)
• Data collected every 30s via chronyc tracking
• Started: 2026-05-23 ~08:19 UTC
• Patched kernel adds: NTP sub-nanosecond precision, NTP precision accounting infrastructure, vmclock_host device. Note: NTP precision accounting is not actively used by chrony in this test — any observable difference comes from the NTP precision patches alone.
• Baseline kernel: same tree minus the 10 NTP precision/vmclock commits

Test Restart — 2026-05-27

Data collection restarted with Amazon Time Sync Service correctly configured (both 169.254.169.123 and fd00:ec2::123 as preferred servers). Previous Fedora 43 default chrony config did not include these; see RHBZ#2482393. IPv6 also enabled on all instances.

Observations (updated at t+3.5h, ~400 samples per host)

• Skew (PPM): Patched hosts show lower skew than their baseline counterparts (A=0.025 vs B=0.046; C=0.072 vs D=0.086 at t+90min). Periodic tick also helps (A < C, B < D). The gap was larger in the first few minutes and has narrowed as all hosts converge.
• RMS Offset: NOHZ_IDLE hosts (C, D) show ~3× higher cumulative RMS (~45µs) than periodic (A, B, ~14µs) over the first 3 hours. Patched is ~10% lower than baseline in both tick modes.
• Offset: All hosts settling to ±10–20µs. Initial transients (first ~10 minutes) dominate the early statistics.
• Allan Deviation (τ=30s): A=60ns, B=174ns, C=325ns, D=408ns. Patched periodic (A) is 3× more stable than baseline periodic (B). NOHZ patched (C) is ~20% better than NOHZ baseline (D). At τ=300s the pattern holds: A=20ns, B=52ns, C=111ns, D=131ns.
• Caveat: These observations are preliminary. Steady-state comparison requires 12+ hours of data after convergence.

Allan Deviation

Allan deviation measures clock stability at different averaging timescales. Lower is better. A clock that's well-disciplined by NTP should show decreasing ADEV at longer timescales (the servo corrects drift). The slope indicates the dominant noise type (white noise = −½, flicker = 0, random walk = +½).

Time Series

The "RMS Offset (chrony)" panel shows chrony's internally-tracked RMS which uses an exponentially-weighted history that predates our test window. The "RMS Offset (cumulative)" panel computes RMS directly from the offset samples collected during this test, giving a fairer comparison unaffected by each host's different convergence history.

Distribution (Box Plot)

CDF of Absolute Offset

Methodology

CSV columns from chronyc -c tracking:

Scripts

• ntp_quality_log.sh — data collector (runs on each host)
• ntp_quality_graph.py — fetch data and generate graphs

Raw Data

• A — periodic + patched
• B — periodic + baseline
• C — nohz + patched
• D — nohz + baseline

CSV Format

Data last fetched: —