Probe traces for "Beyond CPU–GPU Frequency: Memory-Clock and Tail Effects in Edge Inference Latency Estimation"

# Probe traces — jetson-latency-lab (bulk data deposit) Per-pair frequency-transition probe traces for: *"Beyond CPU–GPU Frequency: Memory-Clock and Tail Effects in Edge Inference Latency Estimation"* (Jaehoon Kang). These CSVs (`<domain>_<rateA>_<rateB>.probe.csv`) are the raw per-chunk timing samples behind the RQ3 actuation-lag results (§VI). Each row is `end_ns,dur_ns` on CLOCK_MONOTONIC. They are deposited here because they exceed GitHub's file-size limits; the code that produces and consumes them, and all other (smaller) data, live in the GitHub repository: https://github.com/dankang21/jetson-latency-lab ## Contents | Directory | Run | Role in the paper | |---|---|---| | `pilot_trans/` | final, 150–200 transitions/direction | source of the §VI actuation-lag and stall numbers | | `pilot_trans_run1/` | RT-throttling-contaminated run | the §III-D (P1) pitfall exhibit: kernel RT throttling injecting 46–51 ms phantom stalls | | `pilot_trans_run2_n40/` | intermediate, 40 transitions/direction | superseded by the final run; included for completeness | ## Reproduce Clone the repo, place these directories under `results/`, then: ``` python3 pilot_trans/analyze_trans.py ``` The transition-record CSVs and `meta.json` (already in the repo) are complete without these traces for every analysis except `analyze_trans.py`.