windowed-highcard
Last updated: 2026-06-15 21:38 · same-machine, relative comparison.
Throughput / resource usage
| product | version | offered_in/s | consumed_in/s | output_rate | cpu_avg | cpu_p95 | rss_avg_mb | rss_max_mb | window | run |
|---|---|---|---|---|---|---|---|---|---|---|
| turbine | 0.1.0a4 | 50,000 | - | 3,349.66 | 43.4 | 44.2 | 775.0 | 865.3 | 60 | 2026-06-15 21:38 |
output_rate is the window-emission cadence (≈ keys ÷ window size), not a throughput figure — it only moves when windows close, which needs the consumer to keep up with offered_in/s. Compare CPU/RSS; see the README on benching windowed.
Analysis (turbine = baseline):
cpu_avg(lower better): 🏆 turbinerss_avg_mb(lower better): 🏆 turbine
Sanity — are we benching the same thing?
| product | output_rate | expected | sample |
|---|---|---|---|
| turbine | 3,349.66 | 3,333.00 | 500 |
Aggregated output: checked against the expected pane cadence (expected.output_rate in bench.yaml), not per-record selectivity.
Verdict: ✓ matches theory — output_rate within 2× of the expected pane cadence.