App start time
How long an app takes to open, measured the same way every time, repeated a few times so the median means something.
emulabzip startup com.example.app # cold, 5 runs
emulabzip startup com.example.app --kind hot --runs 10
emulabzip startup com.example.app --json| Kind | Android | iOS Simulator |
|---|---|---|
| cold | The process is stopped first (am force-stop), so the system starts it from nothing | The app is terminated first |
| warm | The screen is closed with the back button and the process is kept | Not available: the simulator has no back button, so there is no separate warm start |
| hot | The app is sent to the background with the home button | The Settings app is put in front |
What is measured#
- Android uses
am start -W, which reportsTotalTime(the time until the first frame is drawn),WaitTimeand theLaunchState. The launcher activity is found withcmd package resolve-activity. - iOS Simulator reads the unified log. SpringBoard logs when it is asked to open an app and again when the app's scene is active. The time between the two is what a person waits, animation included. Compare runs with each other, not with a phone.
The result lists every run and the median, fastest, slowest and 90th percentile. The median is the number to follow; one slow run is usually noise.
Where to find it#
- Desktop: Performance → Start time (kind, number of runs, a bar per run).
- CLI:
emulabzip startup. - MCP:
measure_startup. - CI: with a
[startup]table inemulabzip.toml,emulabzip ci --app …measures before the command and holds the median tostart_time_ms.
The app is stopped and opened on the device while this runs, so do not run it on a device where that costs you work.