smp-comparison.json

{
  "schema_version": "2.0",
  "state": "template",
  "experiment_id": "smp-comparison",
  "title": "ESP32-S3 SMP Comparison",
  "purpose": {
    "question": "How does the same gateway workload behave under ESP-IDF SMP and CONFIG_FREERTOS_UNICORE?",
    "comparison": "Paired physical runs that differ only in the gateway scheduler build and its recorded binary hash.",
    "primary_metric": "Critical on-time delivery alongside gateway queue delay and per-core task evidence."
  },
  "setup": {
    "nodes": null,
    "thread_channel": null,
    "placement": null,
    "firmware_reference": null
  },
  "execution": {
    "warmup_s": null,
    "measurement_s": null,
    "cooldown_s": null,
    "repetitions": null,
    "seed": null
  },
  "traffic": { "streams": null },
  "scenario": { "event": null, "at_s": null, "duration_s": null, "target": null },
  "treatment": { "mode": null, "candidate_action": null, "control_profile": null, "host_model": null, "remote_actuation": null },
  "acceptance": { "counter_reconciliation": null, "minimum_critical_on_time_pdr": null, "negative_case": null },
  "evidence": { "required_artifacts": null, "operator_notes_required": null, "topology_photo_required": null },
  "_todo": [
    {
      "path": "/setup",
      "action": "Build a matched SMP and unicore gateway pair before choosing any workload.",
      "method": "Use one source commit and make only the documented ESP-IDF scheduler configuration differ. Record both sdkconfig files, compiler flags, project task settings, binary hashes, and gateway-specific trace configuration. Keep RCP and endpoint binaries unchanged.",
      "done_when": "A reviewer can identify exactly one intended scheduler difference between the two gateway builds."
    },
    {
      "path": "/execution",
      "action": "Pair scheduler runs within one stable physical block rather than comparing separate days or placements.",
      "method": "Reuse a completed baseline setup and traffic plan, randomly alternate or alternate SMP and unicore runs, reboot the gateway between conditions, and retain only pairs with matching topology, channel, endpoint firmware, and accepted manifest digest.",
      "done_when": "The ready file defines matched repetitions, duration, seed policy, and a predefined exclusion rule for an incomplete pair."
    },
    {
      "path": "/evidence",
      "action": "Require scheduler evidence in addition to network counters.",
      "method": "Archive raw task/core trace, queue delay, stack margin, final counters, versions, and the two scheduler configurations. State a critical service floor and do not interpret an apparent scheduler advantage from runs with unreconciled counters or invalid topology.",
      "done_when": "The report can separate radio variation from a plausible scheduler-related timing change."
    }
  ]
}