Dropbox: Magic Pocket Custom Hardware & Content-Addressable Chunk Storage
How Dropbox stores exabytes of user files by splitting files into 4MB immutable chunks, deduplicating them globally via SHA-256 hashes, and persisting them across custom Magic Pocket storage engines.
Dropbox separates file metadata (file paths, directories, ownership) stored in sharded Edgestore MySQL clusters from binary content stored in Magic Pocket, an exabyte-scale content-addressable storage engine featuring Reed-Solomon erasure coding (8+4).
Content-Addressable 4MB Chunking & Global Deduplication
Exabytes of total storage across billions of user filesStoring millions of duplicate files (e.g. identical PDFs, operating system installers) without wasting storage hardware.
Files are chunked into 4MB blocks, hashed via SHA-256, and stored by hash key. If two users upload the same chunk, only one copy is written to disk.
Chunk boundary calculation requires client CPU hashing overhead, but slashes storage footprints by 30-40%.
In Google Drive or Dropbox interview rounds, always separate the block metadata layer from the raw block chunk storage layer.
Design a File Storage & Sync Service (Dropbox / Google Drive)
Sync files across devices: Chunking (4MB blocks), Content-addressed storage (SHA-256 hash), Delta sync (rsync algorithm), metadata synchronization, and conflict resolution.
Dropbox: Block Storage Sync & Magic Pocket Architecture
Sync exabytes of user files: Client chunking engine, SQLite local metadata, Notification server (long-polling/WebSockets), and custom Magic Pocket hardware block storage.
Wired & Dropbox Engineering • 2020
Ready to Practice Dropbox-Style Systems?
Start with foundational networking, compute, and storage, and build up to complex distributed consensus.