21-EFT.WP.Metrology.Sync v1.0 | Chapter 14 — Quality Indicators, SLOs & Operational Panels

Home ／ Docs-Technical WhitePaper ／ 21-EFT.WP.Metrology.Sync v1.0

Chapter 14 — Quality Indicators, SLOs & Operational Panels

One-line objective: Define synchronization SLIs/SLOs with a unified gauge; build an auditable dashboard and error-budget loop that drives alerting, rollback, and release decisions.

I. Scope & Objects

1. Applies to
   • GM/BC/OC and distributed graph-synchronization paths in PTP/NTP/SyncE/White Rabbit domains.
   • End-to-end metrics and dashboards for ts publication and tau_mono computation chains.
2. Inputs
   • Metric streams: offset(t), skew(t), J(t), asym(t), PDV(t), lock_state(t), holdover_state(t).
   • Arrival fields: dual-form T_arr and delta_form.
   • Windows & thresholds: Delta_t, tol_sync, skew_max, J_rms_max, tol_asym, tol_Tarr, p_target.
3. Outputs
   Time series & quantiles for SLIs, SLO evaluations, error budget and burn rate, panel assets, and exportable manifests manifest.sync.sli.*.

II. Terms & Variables

• Time semantics: ts, tau_mono, Delta_t (sliding window), p_target (target SLO quantile).
• Sync gauges: sync_error(t), skew(t), J_rms, J_pp, asym, PDV.
• Arrival: T_arr, gamma(ell), n_eff, c_ref, delta_form.
• Budget & compliance: bad_time, E_budget, burn_rate, availability, q_score_sync.

III. Axioms P614- **

• P614-1 (Unified measure): Compute all SLIs on tau_mono, publish on ts, and attach offset/skew/J metadata.
• P614-2 (Fixed quantile definition): Use interpolated quantiles; streaming computation must match offline recomputation (error ≤ eps_q).
• P614-3 (Explicit error budget): Express SLOs as bad-time budgets; publish E_budget and burn_rate.
• P614-4 (Dual-form mandatory): Any panel involving arrivals must record both forms and delta_form.
• P614-5 (Dimensional integrity): All expressions pass check_dim(expr) with explicit units (s, s/s, etc.).
• P614-6 (Non-intrusive observation): Instrumentation overhead and probe latency are estimated externally and must not contaminate the measured pipeline.

IV. Minimal Equations S614- **

• S614-1 (Synchronization error)
  sync_error(t) = | ts_ref(t) - ts_local(t) |
  unit "s", dim "[T]"
• S614-2 (Frequency offset & jitter)
  skew(t) = d ts_local / dt - d ts_ref / dt
  J_rms = sqrt( mean_window( ( offset(t) - mean_window(offset) )^2 ) )
  J_pp = max_window(offset) - min_window(offset)
• S614-3 (Asymmetry & PDV)
  asym = | delay_ms - delay_sm |
  PDV = percentile(delay_samples, 99) - percentile(delay_samples, 1)
• S614-4 (Dual-form arrival delta)
  delta_form = | ( 1 / c_ref ) * ( ∫ n_eff d ell ) - ( ∫ ( n_eff / c_ref ) d ell ) |
• S614-5 (Quantile & coverage)
  p_q(x) = inf { v : F_x(v) ≥ q }，coverage = ( observed_points / expected_points )
• S614-6 (SLO decision)
  pass = ( p_target(sync_error) ≤ tol_sync ) ∧ ( p99(|skew|) ≤ skew_max ) ∧ ( J_rms ≤ J_rms_max ) ∧ ( asym ≤ tol_asym ) ∧ ( delta_form ≤ tol_Tarr )
• S614-7 (Error budget & burn)
  bad_time = measure{ t ∈ window : sync_error(t) > tol_sync }
  E_budget = ( 1 - availability_target ) * |window|
  burn_rate = bad_time / E_budget

V. Flow M60-14 (Metrics → Evaluation → Budget → Panel → Persistence)

1. Ready
   • Aggregate raw timestamps and delay samples over Delta_t; normalize units and run check_dim.
   • Bind TraceID/build_id/config_hash and reference-link IDs.
2. Compute SLIs
   • Produce p50/p95/p99, mean, and variance for sync_error/skew/J/asym/PDV; compute delta_form.
   • Compute coverage and observation completeness.
3. Evaluate SLOs
   • Apply S614-6; compute bad_time / E_budget / burn_rate and trend.
   • Generate q_score_sync ∈ [0,1] (e.g., q = max( 0, 1 - burn_rate ), then apply multi-metric weighting).
4. Panels & alerts
   • Render panel spec: time series, quantile bands, budget bars, alert thresholds, rollback hooks.
   • Alert tiers: warn (near threshold) → major (threshold breach ×1) → critical (burn_rate > br_max).
5. Persist & sign
   • Export manifest.sync.sli.*, contract_report, panel_spec.json; sign with hash_sha256(blob) and signature.
   • Failure path: block freeze, trigger rollback and re-verification window.

VI. Contracts & Assertions (C60-14x)

• C60-140 (Dimensional integrity): Run check_dim(expr) across all S614-* expressions.
• C60-141 (Coverage): coverage ≥ 0.99; otherwise mark window invalid and de-weight.
• C60-142 (Sync quantile): p_target(sync_error) ≤ tol_sync.
• C60-143 (Skew ceiling): p99(|skew|) ≤ skew_max.
• C60-144 (RMS jitter): J_rms ≤ J_rms_max; if J_pp > J_pp_max, initiate link diagnostics.
• C60-145 (Asymmetry): asym ≤ tol_asym.
• C60-146 (Dual-form arrivals): delta_form ≤ tol_Tarr.
• C60-147 (Budget burn): burn_rate ≤ br_max (e.g., br_max = 1.0).
• C60-148 (Panel freshness): now - last_panel_emit ≤ freshness_slo.
• C60-149 (Lock times): t_lock ≤ t_lock_max; after unlock t_relock ≤ t_relock_max.
• C60-14A (BMCA stability): gm_switch_count per unit time ≤ gm_switch_max.

VII. Implementation Bindings I60-14*

• compute_sync_sli(stream, window, qset) -> {p50,p95,p99,mean,var}
• compute_jitter_and_pdv(delay_samples) -> {J_rms,J_pp,PDV}
• calc_arrival_two_forms(path, n_eff, c_ref) -> {T_arr_1,T_arr_2,delta_form}
• evaluate_slo(sli, rules) -> {pass, violations, bad_time, E_budget, burn_rate, q_score_sync}
• emit_dashboard(spec, series) -> panel_id
• export_sli_manifest(ctx, sli, slo) -> manifest.sync.sli
• sign_panel(blob, key_id) -> {hash_sha256, signature}
• Invariants: streaming quantiles match offline recomputation; manifest.sync.sli is replayable; delta_form and thresholds share a common config source.

VIII. Cross-References

• Servo & filtering: Chapter 6.
• Delay & asymmetry: Chapter 7.
• Offset/frequency estimation: Chapter 8.
• Noise & stability: Chapter 9 (interpretation of J and Allan family).
• Compliance & audit: Chapter 13.
• Arrival & path consistency: EFT.WP.Metrology.TimeBase v1.0 Chapter 9.
• Cleaning, release & manifests: Methods.Cleaning v1.0 Chapter 10.
• Statistical SLOs & error budgets: Methods.CrossStats v1.0 Chapter 14.

IX. Quality Layers & Risk Control

1. Key dashboard layers
   sync_error quantile bands (p50/p95/p99) with threshold lines; cumulative burn_rate bars; J_rms / J_pp and PDV trends; daily asym; delta_form trajectories.
2. Risk actions
   • Mild: de-weight offending paths, lengthen servo constants, raise packet-rate limits.
   • Moderate: switch sources or escalate upstream class; trigger path symmetry re-measurement.
   • Severe: enter holdover & traffic isolation; freeze release and mandate audit.

Summary