Instrumented HID trial: DPC evidence
A separate trial used WPR GeneralProfile file mode in the dedicated Windows guest. The producer generated 100 warm-up and 1,000 measured vendor-HID reports at a requested 1,000 reports/s. Its purpose is instrumented, so it is excluded from primary latency charts.
| Recorded item | Value |
|---|---|
| Whole trace duration (seconds) | 28.206191 |
| Lost events | 0 |
| Lost buffers | 0 |
| Vdevice module DPC calls | 2221 |
| CPU 0 Vdevice DPC duration (microseconds) | 20476 |
| CPU 1 Vdevice DPC duration (microseconds) | 25757 |
These are module-attributed DPC durations from xperf -a dpcisr, across the whole trace, including enumeration, warm-up, report generation, waiting and teardown. They are not total driver CPU, time per report, proof of ISR behavior or a competitor ranking. The instrumented trial and uncontrolled shared-host scheduling cannot be merged with the uninstrumented headline latency results.
Reproduce and inspect
Use automation/trace-scenario.ps1 only after restoring the owned Vdevice checkpoint. The script starts its own WPR recording and stops it without cancelling someone else's existing trace. scripts/summarize_etw.py decodes the local ETL and exports only aggregate measurements. The trace remains local because it contains broad guest system state.
ETL SHA-256: 7946d0ad8c70c1751c2171e53043e54c1e0769bded583173bc0375b69a876bdc. The raw archive includes reports/etw-validation.json; the ETL itself is excluded.