Greenhouse imagery and deployment telemetry for a containerised edge vision node on a Raspberry Pi 5

Version 2.0.0 of this record holds the three-condition measurement campaign supporting the study "Container Overhead or Stale Dependencies? Decomposing the Latency and Energy Cost of Containerised Inference on an Edge Node": nine three-hour runs (27-28 September 2026) on a Raspberry Pi 5 comparing native execution, a container with the host's interpreter and C library versions and byte-identical compiled numerical modules, and a legacy container (Python 3.9, ONNX Runtime 1.19.2), in a randomised complete block design. Node input power, battery current, CPU frequency and cap, under-voltage alarm and fan speed are logged at 1 Hz through both the active and the sleep phase of every duty cycle. The archive includes per-frame and 1 Hz logs, cycle events, run summaries, provenance records, camera settings, environment fingerprints, the orchestration log and the run kit; see DEPOSIT_README.md and the README.md inside the archive.

The greenhouse imagery and the first campaign's telemetry are not repeated here; they remain in version 1.0.0 of this record: https://doi.org/10.5281/zenodo.22857312

Analysis code that regenerates every number, table and figure of the study: https://github.com/xiaolin200206/edge-container-deployment-measurement (directory three_condition, release v2.0-tsusc).

Licence: CC BY 4.0.