Testing SDK Startup Time: Cold, Warm, and Lukewarm Launches on Android

The author, responsible for testing a commercial advertising SDK, discusses the need to evaluate app launch time because media partners often report performance issues that affect user experience.

Android launch types are introduced: Cold Start (app process not running), Warm Start (activity brought to foreground with retained process), Lukewarm Start (partial cold‑start operations), and Warm Start in Chinese terminology (including "温启动"). Definitions are taken from the Android Developers documentation.

For the SDK under test, three launch scenarios are selected—first cold start, non‑first cold start, and warm start—because they differ in cloud‑control requests, configuration fetching, and ad initialization, and they minimize demo impact on measured times.

The testing environment includes using a pure Android device (e.g., Nexus), consistent network conditions (Wi‑Fi, cellular, weak), single‑ vs multi‑process modes, whether splash‑ad pre‑loading is enabled, and the size of ad media that may compete for resources.

Launch time is measured with the ADB command adb shell am –W [packageName]/[packageName.MainActivity] , which returns ThisTime, TotalTime, and WaitTime; TotalTime is used as the primary metric for SDK startup performance.

An example shell snippet for capturing TotalTime is shown: adb shell am start -W $PackageName/$ActivityName | grep TotalTime | awk -F ' ' '{print $2}' | tr -d "\r"

To reduce variability, tests should be repeated, outlier values (highest and lowest) discarded, and the average taken, while ensuring no other apps interfere with the measurement.

The author also reports a problem with the awk syntax in the above pipeline, explaining that the command failed to retrieve data and directing readers to an online awk tutorial for correction.