Mobile App Performance Metrics

The Performance Score (PS) which is used in Google Lighthouse is summing up the speed index, the first paint time, the layout load time, the CPU stable time and the rendering time after data download metrics in percentile on their distribution with coefficients as follows.
The Speed index (SI) provides an insight on how fast the mobile app is displayed on the screen. The speed Index in WebPageTest is calculated as the accumulated remaining area that has not been rendered or visually complete.
First paint time (FP): After a user event occurred, the mobile app will change the display. We record the mobile app video file, and compare every image snapshot to find the first paint image with the bitmap difference compared with the previous one.
Layout load time (LL): While analyzing the xmldump file, we can know when the mobile app has completed the layout load by observing the layout count.
CPU stable time (CS): When a mobile app runs on the smartphone, it consumes the CPU cycles. Hence, the CPU usage increases for short time and returns to the stable point. We capture the CPU stable time after the user input.
Rendering time after data download (RT): A mobile app usually downloads contents from the server and renders its data on the display. As a mobile app has many objects and contacts to multiple servers, we compute the rendering time after the traffic download is completed.

PS = 0.6*SI + 0.1*FP + 0.1*LL + 0.1*CS + 0.1*RT

Mobile app speed index

The speed index provides an insight on how fast the mobile app is displayed on the screen. The speed index in the WebPageTest is calculated as the accumulated remaining area that has not been rendered or visually complete up to the session end time of the test user event. The low speed index means the quick response of rendering the display after receiving the traffic from the server or executing the embedded content within the app.
Though the speed index can be applied to compute the speed index of a mobile app, AppSpeedXray have to solve how to detect the session end when testing a user input of touches or drags.

Since it is not possible to find rendering completion event time directly from a mobile app, AppSpeedXray record the video file of a mobile app during the test, and determine the rendering completion event time when the similarity metric between two images becomes larger than the threshold.
- Rendering completion detector look for a candidate XML layout time window which is recorded by xmldump every two to four seconds(The time granularity depends on device CPU computing power.). An XML layout time window is considered as the candidate time window including the rendering completion time event, if its layout count is constant compared with the previous window 5 times in a row.
- Rendering completion detector identify the specific rendering completion image by computing the structural difference of images extracted every 100 ms within the candidate XML layout time window(Basically it use SSIM for the structural difference algorithm with the similarity threshold (e.g., 90\%)).

Report abuse