smoke

Smoke Example Outputs

This directory contains one comparable 60-second run for each supported bandwidth estimator:

• gcc/
• gemini/
• hrcc/

All three runs use:

• media: benchmark/media/Gameplay.mp4
• trace: benchmark/traces/fcc_wried/251_HIGH_STABLE.json
• duration: 60 seconds

Each result directory contains the emulator logs, telemetry, and a derived qoe.json score. The media, trace, output_dir, and workdir fields in run.json are rewritten as repository-relative paths after the run so the archived examples remain portable.

The QoE score and its normalized components are:

Model	QoE (0-100)	Bitrate score	Jitter score	Freeze score
GCC	81.644	0.645	0.820	0.985
Gemini	52.657	0.146	0.453	0.981
HRCC	59.032	0.258	0.513	1.000

The receiver telemetry underlying those scores is:

Model	Received bitrate (Kbit/s)	Jitter-buffer delay (ms)	Freeze duration (ms)	Frames decoded	Dropped frames	Render FPS
GCC	1306	27	916	3536	6	59
Gemini	295	82	1153	2417	5	40
HRCC	522	73	0	3170	4	53

Run the Example

Run from any directory:

./examples/smoke/run_examples.sh

New results are written to the ignored results/example-runs/smoke/ directory, so the archived reference outputs under examples/smoke/ remain unchanged. Set OUTPUT_ROOT to use a different output location:

OUTPUT_ROOT=/tmp/onl-examples ./examples/smoke/run_examples.sh

The script refuses to overwrite non-empty result directories. It runs the models sequentially because all cases share the loopback tc configuration. After each case, it waits five seconds and verifies that the temporary workdir is unmounted and empty or removed. Pass model names to run only a subset:

./examples/smoke/run_examples.sh hrcc

The run requires the same Linux, Git LFS, FFmpeg, tc, tmpfs, and sudo dependencies documented in the repository root README.

Recalculate QoE

python3 benchmark/calculate_qoe.py examples/smoke/gcc
python3 benchmark/calculate_qoe.py examples/smoke/gemini
python3 benchmark/calculate_qoe.py examples/smoke/hrcc