Currently, frame times percentiles are treated as individual metrics, with their own internal distributions:
frameTime50thPercentileMs min 6.0, median 6.0, max 7.0
frameTime90thPercentileMs min 7.0, median 7.0, max 9.0
frameTime95thPercentileMs min 7.0, median 7.0, max 9.0
frameTime99thPercentileMs min 14.0, median 14.0, max 14.0
totalFrameCount min 294.0, median 294.0, max 294.0
Instead, we should flatmap all metrics captured from separate iterations, and compute metrics from those, e.g:
frameTimeMs P50 7.1, P90 9.4, P95 9.6, P99 12.0
In order to match this treatment of data in JSON, we should surface percentiles without min/max/median, while still preserving per-iteration separation of data, to enable arbitrary post-processing it.
Description
Currently, frame times percentiles are treated as individual metrics, with their own internal distributions:
Instead, we should flatmap all metrics captured from separate iterations, and compute metrics from those, e.g:
In order to match this treatment of data in JSON, we should surface percentiles without min/max/median, while still preserving per-iteration separation of data, to enable arbitrary post-processing it.