AIW: A Free Android Performance Ebook That Pins Every Claim to AOSP Source
A performance reference that maps each claim to a specific AOSP tag gives Android engineers a checkable alternative to generic advice, but the Android 17 baseline means readers still have to verify findings against their own OEM kernels and device behavior.
AIW organizes Android internals into five parts: system mechanisms (architecture, rendering, input, memory, CPU scheduling, storage I/O), performance issues (jank, ANR, memory, power, network, render pipeline), tools and methodology (Perfetto, APM), vendor/OEM differences, and app-level practices. Each mechanism section cites AOSP classes or file paths, with the current platform baseline pinned to android-17.0.0_r1 and the kernel baseline to android17-6.18-2026-06_r6.
The book deliberately flattens the usual fragmentation problem. It pairs symptoms with concrete code paths and tooling: a slow start is broken into process creation, initialization, first frame, and business-available timestamps, each tied to chapters that can be cross-read with Perfetto. It warns that rendering activity doesn't mean a frame is displayed, and that low CPU usage doesn't rule out thread queues. The EPUB is rebuilt weekly on GitHub Releases, and a separate Knowledge Pack makes the 26 chapters searchable with version tags for use inside SmartPerfetto. Content is licensed CC BY-NC-SA 4.0, with commercial use requiring separate permission.
Pinning the entire book to one AOSP tag makes the advice checkable but also dated: by the time Android 17 devices are widespread, the next version's code paths will have diverged, so readers still need to verify against their own kernels.
The repeated examples of symptoms not mapping to causes (rendering activity without a displayed frame, low CPU with queued threads) point to a core truth: Android performance debugging is about correlating asynchronous stages, not reading a single metric.
Inviting readers to correct the text via Issues and PRs is a verification loop that most published performance books lack, and it aligns with the warning that correlation is not causation.
The book's structure assumes readers already have Android development or performance testing experience; it does not aim to teach first-time app developers.