Baidu APP Android Live Streaming Startup Performance Optimization Practice

This article details the optimization practice for live streaming playback startup time in Baidu APP Android version. Due to plugin-based architecture for package size management, the live streaming function uses a two-level plugin system: a smaller level-1 plugin with core capabilities (playback, IM) built-in, and a larger level-2 plugin for business features (widgets, gifts, likes). This special plugin logic and complex business scenarios caused poor startup performance metrics.

Background and Goals: In Q1 2022, the overall 80th percentile startup time was around 3 seconds, with second-entry (swiping within live room) at about 1 second. After plugin splitting, the first-entry scenario with stream address grew from 1.5s to 2.5s+ within two weeks. The target was set to 1.5 seconds for general scenarios and 1.1 seconds when stream address was pre-obtained.

Key Solutions Implemented:

1. New Monitoring Reports: Designed custom reporting to measure time between any two key nodes in the startup chain, enabling precise identification of time-consuming segments.

2. Level-1 Plugin for First Entry: Modified first-entry logic to use level-1 plugin entirely, reducing multi-layer call overhead from level-2 plugin. Also implemented a live room data request manager to parallelize interface requests with page creation.

3. Pre-starting Player Outside Activity: Created a pre-startup module that initiates player creation and buffering in parallel when user clicks, before the Activity launches. This achieved 450ms+ optimization in 80th percentile startup time.

4. Task Scattering for Live Room: Split large component initialization tasks into smaller individual tasks to avoid blocking the main thread, reducing level-2 plugin component loading time by 40%+.

5. Kernel Sub-thread First Frame Distribution: Implemented sub-thread notification for first frame events, using postAtFrontOfQueue() to prioritize first frame processing, reducing first frame distribution delay.

6. Balancing Startup and Full-load Metrics: Optimized plugin pre-download and component initialization prioritization to improve full-load metrics without degrading startup performance.

Results: After optimization, the overall startup time decreased from 3s+ to approximately 1.3s; first-entry with stream address dropped from 2.5s+ to under 1s; second-entry reduced from 1s+ to under 700ms.