Home / Docs-Technical WhitePaper / 27-EFT.WP.Packets.Light v1.0
Chapter 14 — Runtime & Streaming (Caching / Fallback / Panels)
One-sentence goal: Build a streaming-runtime convention for optical-packet links on an execution graph G = (V, E) centered on windows / caching / fallback / panels, ensuring dual-form compliance (configured vs measured) and dimensional correctness, with runtime evidence published via a manifest.
I. Scope & Objects
- Inputs
- Framing & timebase: frame_spec = { T_epoch, T_f, T_slot, T_sym, T_guard } (Ch. 3), monotone settlement clock tau_mono and publication instant ts.
- Runtime configuration: window width/stride W / H, allowed lateness lateness, watermark policy watermark, cache policy cache.policy (ttl / size / key), rate-limit / backpressure policies.
- Upstream manifests: artifacts from physics / compensation / metrology / harmonization—manifest.packet.{ phys, comp, meas, tarr, swrt, … }.
- Outputs
- Online products: windowed metrics and arrival time T_arr*, metrology (OSNR / EVM / BER), switching/queueing delays lat_*, and the online dual-form error delta_form_rt;
- Runtime panels & alerts: panel.* (SLI/SLO curves, quantiles/heatmaps), fallback traces;
- Runtime manifest: manifest.packet.rt.* (defined in this chapter).
- Boundary
Physics & compensation: Chs. 2 / 6; switching & queueing: Chs. 9 / 10. This chapter focuses on execution graph / caching / fallback / panels.
II. Terms & Variables
- Windows & lateness: win_k = [ t_k, t_k + W ), hop H, overlap O = 1 − H/W, allowed lateness lateness; watermark function wm(t).
- Cache: key K_cache = hash( frame.hash, path.hash, label, RefCond, algo.hash, win_id ); hit rate hit; staleness stale_ratio.
- Rate-limit & backpressure: input rate λ_in, service rate μ, queue depth q, drop rate drop_rate; token bucket ( r, B ).
- Dual forms (runtime): configured (model/manifest expected values, e.g., lat_total^{config}, T_arr^{config}) vs measured (windowed timestamps / metrology), discrepancy delta_form_rt.
- Panels & resources: latency_p{q}, cpu_pct, mem_pct, gpu_util, err_rate, coverage.
- Dimensions: unit(W, H, lateness, latency) = "[T]", unit(hit, stale_ratio, drop_rate, coverage) = 1.
III. Postulates P614-*
- P614-1 (Two forms in parallel): For each window win_k, compute and publish metric^{config} and metric^{meas}, recording delta_form_rt = | metric^{config} − metric^{meas} |.
- P614-2 (Temporal semantics & window close): A window closes if wm(t_proc) ≥ t_k + W − lateness; late data triggers patch recompute or a shadow pipeline—policy must be persisted.
- P614-3 (Traceable cache): Cache keys include algorithm and RefCond; any hash change invalidates and recomputes.
- P614-4 (Dimensional compliance): Run check_dim( y − f(x) ) before publication; log any log↔linear conversions in manifest.packet.rt.*.
- P614-5 (Fail-Closed fallback): On SLO/contract or dual-form assertion failure, degrade / bypass / rollback to the last signed version and persist the strategy card.
- P614-6 (Privacy & minimum visibility): Runtime manifests contain no raw sensitive data—only hashes/pointers/statistics/signatures.
IV. Minimal Equations S614-*
- Window fusion & publication
- Weighted fusion (example: online arrival-time harmonization):
T_arr*,win = w_cnt t̂_cnt + w_cont t̂_cont + w_phys T_form2, with weights w_* ∝ 1 / u_*^2 (uncertainties from the most recent window). - Published quantiles:
z_{p95} = quantile( { z_i | i ∈ win_k }, 0.95 ).
- Weighted fusion (example: online arrival-time harmonization):
- Online dual-form error
delta_form_rt(metric) = | metric^{config}(win_k) − metric^{meas}(win_k) |, for metric ∈ { T_arr, OSNR, EVM, lat_total, P_block, P_cont }. - Cache & staleness
- Che approximation (hit): find T_c s.t. ∑_u ( 1 − e^{−λ_u T_c} ) = cache.size; then hit_u ≈ 1 − e^{−λ_u T_c},
hit = ( ∑ λ_u hit_u ) / ( ∑ λ_u ). - Staleness: stale_ratio = | items_expired_but_used | / | items_used |; must satisfy ≤ tol_stale.
- Che approximation (hit): find T_c s.t. ∑_u ( 1 − e^{−λ_u T_c} ) = cache.size; then hit_u ≈ 1 − e^{−λ_u T_c},
- Rate-limit / backpressure
- Token bucket: generation r, bucket B; instantaneous burst ≤ B, long-term average ≤ r; choose r = ρ • μ (ρ < 1).
- Stability: E[λ_in] < E[μ]; queue recursion q_{k+1} = max( 0, q_k + a_k − s_k ).
- SLO breakdown
- End-to-end delay: lat_total = lat_prop + lat_ser + lat_sw + lat_q + lat_proc (per Ch. 10);
- Target: p95(lat_total) ≤ SLO.latency_p95, drop_rate ≤ SLO.drop_max.
- Fallback ladder (expression)
mode := scale_down → damping → lower_order → bypass → rollback_version, each step persisting actions[] and outcome metrics.
V. Metrology Pipeline M60-14 (Ready → Execute → Verify → Fallback → Persist)
- Ready: freeze frame_spec, RefCond, W/H/lateness/watermark, cache.policy, and SLO/contract thresholds; load upstream manifests and algorithm signatures.
- Execute: advance windows by H; look up K_cache; on miss, compute; log latency, cpu/mem/gpu, hit, stale_ratio, watermark_lag.
- Verify:
- Dual-form errors delta_form_rt;
- SLO/contracts: latency_p95 / drop_rate / hit / stale_ratio and ρ / queue depth;
- Dimensional & mask/band consistency (aligned with Chs. 4 / 11).
- Fallback: when any key assertion fails, execute strategy cards per S614-6; persist actions, blast radius, and recovery.
- Persist:
manifest.packet.rt.* = { frame.hash, path.hash, algo.hash, RefCond, W, H, lateness, watermark, cache:{ ttl, size, policy, hit, stale_ratio }, metrics:{ delta_form_rt, latency_p{50,95,99}, drop_rate, ρ, watermark_lag }, resources:{ cpu_pct, mem_pct, gpu_util }, actions[], contracts.*, signature }.
VI. Contracts & Assertions C60-14x (Suggested Thresholds)
- C60-1401 (Dual-form gap): delta_form_rt_p95 ≤ tol_rt (business-specific).
- C60-1402 (Latency / jitter): latency_p95 ≤ SLO.latency_p95, jitter_p95 ≤ 0.2 • W.
- C60-1403 (Cache efficacy): hit ≥ hit_min (typ. ≥ 0.8), stale_ratio ≤ tol_stale.
- C60-1404 (Backpressure stability): E[λ_in] < E[μ] and drop_rate ≤ drop_max.
- C60-1405 (Panel freshness): panel.update_interval ≤ Δt_panel_max, coverage ≥ cov_min.
- C60-1406 (Dimensional compliance): time / rate / probability fields pass check_dim; dB↔linear conversions recorded.
VII. Implementation Bindings I60-14* (interfaces, I/O, invariants)
- I60-141 plan_stream_windows(W, H, lateness, watermark) -> schedule
- I60-142 query_or_compute(cache, K_cache, fn) -> { value, hit, stale }
- I60-143 enforce_rate(token_bucket:{ r, B }, now) -> { permit, state }
- I60-144 measure_runtime(stream) -> { latency, drop_rate, ρ, cpu_pct, mem_pct, gpu_util, watermark_lag }
- I60-145 compare_dual_runtime(cfg_metrics, meas_metrics) -> delta_form_rt
- I60-146 apply_fallback(mode, ctx) -> actions[]
- I60-147 snapshot_panel(metrics, resources, u/U) -> { panel.uri, snapshot.hash }
- I60-148 emit_runtime_manifest(results, policy) -> manifest.packet.rt
Invariants: two_forms_present = true; schedule is monotone; cache keys include algorithm/convention and RefCond; every fallback is replayable and auditable.
VIII. Cross-References
- Framing & conservation: Chapter 3.
- Physics / compensation / metrology: Chs. 2 / 6 / 11.
- Arrival-time harmonization: Chapter 8.
- Switching / queueing: Chs. 9 / 10 (lat_*, resources & conservation).
- Security & integrity: Chapter 13 (alert linkage and mitigation).
IX. Quality & Risk Control
- SLI / SLO: delta_form_rt_p95, latency_p95, drop_rate, hit, stale_ratio, ρ_p95, watermark_lag_p95, panel_freshness.
- Fallback gradient: scale_down → damping → lower_order → bypass → rollback_version, coordinated with security/integrity policies (Ch. 13).
- Audit: window signatures & cache keys, SLO breaches and dual-form distributions, fallback actions & impact scope, and the manifest.packet.rt.* signature chain with replay consistency.
Summary
- This chapter establishes the streaming runtime baseline for optical packets: windowing, cache reuse, rate-limit/backpressure, fallback, and panels.
- Using P614 / S614 / M60-14 / C60-14x / I60-14* and manifest.packet.rt.*, runtime becomes traceable, auditable, and rollback-ready, interlocking with prior chapters’ conventions across physics / compensation / metrology / routing / queueing.
Copyright & License (CC BY 4.0)
Copyright: Unless otherwise noted, the copyright of “Energy Filament Theory” (text, charts, illustrations, symbols, and formulas) belongs to the author “Guanglin Tu”.
License: This work is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0). You may copy, redistribute, excerpt, adapt, and share for commercial or non‑commercial purposes with proper attribution.
Suggested attribution: Author: “Guanglin Tu”; Work: “Energy Filament Theory”; Source: energyfilament.org; License: CC BY 4.0.
First published: 2025-11-11|Current version:v5.1
License link:https://creativecommons.org/licenses/by/4.0/