System Design Problems

Design Dropbox

Dropbox serves 500M+ users with 2B+ files synced daily. This design covers file chunking, real-time synchronization, conflict resolution, and collaborative editing.

Scale — 500M+ users, 2B+ files, 10M+ concurrent
Sync — Real-time with < 1 second propagation
Storage — Exabytes of user files with deduplication

Dropbox's core challenge is maintaining file consistency across millions of devices while handling conflicts gracefully.

Requirements Clarification

Functional Requirements

Upload and download files
Automatic file synchronization across devices
File versioning and history
Share files and folders with others
Offline access with sync on reconnect
Conflict detection and resolution

Non-Functional Requirements

Availability: 99.99% uptime
Durability: 99.999999999% (11 nines)
Consistency: Strong for file metadata, eventual for content
Scale: 500M users, 10M concurrent

Dropbox's key insight: Block-level sync. Instead of syncing entire files, Dropbox splits files into 4MB blocks and only syncs changed blocks. This reduces bandwidth by 10-100x.

Back-of-the-Envelope Estimation

Storage per User

\text{Storage} = 2000 \text{ files} \times 1 \text{ MB avg} = 2 \text{ GB/user}

Here,

$2000$ =Average files per user
$1 MB$ =Average file size

Total Storage Estimation

500M users x 2 GB = 1 EB (exabyte)
With 3x replication: 3 EB
With deduplication (40% savings): 1.8 EB

High-Level Architecture

File Chunking and Deduplication

DfContent-Addressable Storage

Dropbox splits files into 4MB blocks and hashes each block using SHA-256. The hash becomes the block's identifier. If two users have the same block, it's stored once (deduplication). This reduces storage by 40-60%.

Block Hash

\text{block\_id} = \text{SHA-256}(\text{block\_data})

Here,

$block_data$ =4MB chunk of file data
$block_id$ =Unique content hash

Dropbox uses a Merkle tree to efficiently detect which blocks changed. The tree hash of a file changes only when a block changes, enabling O(log n) change detection.

Synchronization Protocol

Conflict Resolution

DfLast-Writer-Wins with Manual Override

Dropbox uses last-writer-wins (LWW) for non-conflicting changes. When two users edit the same block simultaneously, Dropbox creates a conflict copy (filename.conflict). Users manually merge conflicts.

Conflict Detection

\text{Conflict} = (\text{version}_A \neq \text{parent}) \wedge (\text{version}_B \neq \text{parent})

Here,

$version_A$ =Device A's version
$version_B$ =Device B's version
$parent$ =Common ancestor version

Data Model

File Metadata Schema

\text{File} = (file\_id, name, path, size, blocks[], version, modified\_at)

Here,

$file_id$ =Unique file identifier
$blocks[]$ =Ordered list of block hashes
$version$ =Monotonically increasing version

Practice Exercises

Sync Design: How would you handle syncing a 10GB file where only 1 byte changed?
Conflict: Design a conflict resolution system for collaborative document editing.
Offline: How would you handle a user editing files offline for a week, then reconnecting?
Storage: Design a deduplication system for 1 EB of user files.

Key Takeaways:

Block-level sync with content-addressable storage enables deduplication
Merkle trees enable efficient change detection
LWW with conflict copies for concurrent edits
WebSocket push for real-time sync notifications
4MB block size balances granularity vs overhead

What to Learn Next

-> Design Google Search Web-scale indexing and search.

-> Design WhatsApp Messaging and real-time delivery.

-> Idempotency Handling duplicate requests safely.

-> Outbox Pattern Reliable event publishing.

-> Strangler Fig Incremental migration strategies.

-> Retry Patterns Resilient retry mechanisms.

Design Dropbox

Design Dropbox

Requirements Clarification

Functional Requirements

Non-Functional Requirements

Back-of-the-Envelope Estimation

Storage per User

Total Storage Estimation

High-Level Architecture

File Chunking and Deduplication

DfContent-Addressable Storage

Block Hash

Synchronization Protocol

Conflict Resolution

DfLast-Writer-Wins with Manual Override

Conflict Detection

Data Model

File Metadata Schema

Practice Exercises

What to Learn Next

Premium Content

Need Expert System Design Help?