TechPrufer
Get started
Meta
Meta logoMeta · Feature trail

App Size Compression and Optimization

1 post · Nov 2013 – Sep 2021

What this trail covers

Meta's approach to managing app size has evolved to include advanced compression techniques beyond traditional methods. Project LightSpeed for Messenger on iOS involved a complete rewrite of the codebase, reducing core code by 84% and binary size by 75%. This was achieved by leveraging native OS features, reusing UI components with dynamic templates powered by SQLite, and using SQLite as a universal system for data storage and management. A server broker was introduced as a gateway to server features. This post details efforts to improve Messenger's startup time and overall performance on Android and iOS by removing non-essential code, optimizing dependency injection via bytecode manipulation (ASM), reusing view controllers, moving audio playback to background threads, and reducing network requests. It also highlights the implementation of automated startup-timing tests on physical devices to catch performance regressions.

All Meta trails
  1. Superpack: Pushing the limits of compression in Facebook’s mobile appsThis post introduces Superpack, a new technique for compressing mobile application binaries. Superpack combines compiler analysis (parsing, code generation, SMT solvers) with data compression techniques (LZ parsing, entropy coding) to achieve higher compression ratios than traditional methods. It improves LZ parsing by grouping data based on AST and exploiting semantic knowledge. It enhances entropy coding by identifying contexts with lower entropy for literals. Superpack also transforms unstructured data into programs for compression. It is applied to Dex bytecode, ARM machine code, and Hermes bytecode, resulting in over 20% average size reduction for Meta's Android apps.
  2. Project LightSpeed: Rewriting the Messenger codebase for a faster, smaller, and simpler messaging appThis post details Project LightSpeed, a complete rewrite of the Messenger iOS codebase. Key technical contributions include: reducing core code from 1.7M lines to 360,000, resulting in an 84% reduction; achieving a 75% reduction in app size (from >130MB to one-fourth the size); doubling startup speed; using native OS features for rendering, transcoding, and threading; reusing UI components via dynamic templates driven by SQLite; and implementing SQLite as a universal system for data storage, concurrency, caching, and transactions. A server broker was introduced as a universal gateway between the app and server features.
  3. Facebook open-sources Spectrum 1.0.0 for better mobile image productionMeta has open-sourced Spectrum 1.0.0, a cross-platform image processing library for Android and iOS. Spectrum addresses the challenges of large mobile image file uploads by providing efficient transcoding, resizing, and format conversion. It features a C++ core for consistency and performance, with Java and Objective-C APIs. Spectrum integrates with advanced compression libraries like MozJpeg and offers a declarative API with 'recipes' and plugins to optimize image operations, reducing upload file sizes by up to 15% and improving reliability and quality.
  4. 5 ways Facebook improved compression at scale with ZstandardThis post details the adoption and enhancement of Zstandard (zstd) across various Facebook systems, including large payload backups, package distribution, data warehousing, compressed filesystems (SquashFS, Btrfs), and databases (RocksDB-based systems). Key advancements include multithreaded compression (`-T#`), long-range mode (`--long`) for larger window sizes (up to 2GB), automatic level determination (`--adapt`), increased block sizes in columnar stores (beyond 256KB), reduced data transformations that previously aided zlib, and faster decompression speeds for SquashFS. These optimizations have led to significant improvements in compression ratios (e.g., 16-27% for backups, 10% for fbpkg, 5x in extreme cases, 15% smaller HHVM packages), decompression speeds (2x faster reads for HHVM), and overall storage efficiency (3x more storage on Btrfs dev servers).
  5. Automatic memory leak detection on iOSIntroduced FBRetainCycleDetector for automated retain cycle detection in Objective-C, FBAllocationTracker for efficient instance tracking, and FBMemoryProfiler for manual profiling with generation tracking. Detailed the use of Objective-C runtime introspection (ivar layouts, type encoding) and block ABI analysis to identify strong references causing retain cycles. Described a backend algorithm for clustering retain cycles and using 'git blame' to assign fixes. Outlined generation tracking in FBMemoryProfiler for identifying long-lived objects.
  7. Open-sourcing ReDex: Making Android apps smaller and fasterThis post announces the open-sourcing of ReDex, an Android bytecode optimizer. It details new optimizations implemented since its previous announcement, including feedback-directed class layout for improved startup performance, removal of single-implementation interfaces to reduce method count and improve call efficiency, and stripping of unnecessary metadata like source file names and annotations. The post also provides performance numbers showing significant reductions in app size (25%) and cold start times (20-30%) on real-world devices. It highlights ReDex's design principles for developers: speed, simple interface, and configurability, and encourages community contributions.
  8. Optimizing Android bytecode with ReDexIntroduced Redex, a pipeline for optimizing Android .dex files. Redex applies customizable transformations at the bytecode level for global, inter-class optimizations. Specific optimizations discussed include minification of human-readable strings (with back-mapping for source file paths), inlining of wrapper and detour functions to reduce overhead and bytecode, and dead code elimination by marking reachable functions starting from known entry points. The pipeline is architected as a series of independent optimization plugins for flexibility and parallel development.
  9. Reducing FOOMs in the Facebook iOS appThis post details the identification and mitigation of Out-Of-Memory (OOM) crashes in the Facebook iOS app. It describes the process of distinguishing FOOMs (foreground OOMs) from native crashes, developing a logging mechanism to track app termination reasons, and profiling memory usage with Instruments. Key contributions include the migration from UIWebView to WKWebView to offload memory usage to a separate process, optimizing temporary memory allocations, and implementing an in-app memory profiler using Objective-C method swizzling to track object allocations and detect leaks. The post also mentions leveraging CT-Scan infrastructure for flagging memory regressions.
  10. Building a faster MessengerThis post details efforts to improve Messenger's startup time and overall performance on Android and iOS by removing non-essential code, optimizing dependency injection via bytecode manipulation (ASM), reusing view controllers, moving audio playback to background threads, and reducing network requests. It also highlights the implementation of automated startup-timing tests on physical devices to catch performance regressions.