Designing Twitter/X at Scale

Twitter serves 500M daily users posting 500M tweets per day.

Core Features

• Post tweets (280 chars, media)
• Follow/Unfollow users
• Home timeline
• Trending topics
• Search

Data Estimations

Tweets/day: 500M
Read:Write ratio = 100:1
Peak reads: ~600K/sec
Storage: ~500GB/day (text only)

Tweet Storage

• Tweets: MySQL (strong consistency for writes)
• Media: S3 + CDN
• Timeline Cache: Redis sorted sets

Timeline Architecture

Hybrid fan-out:

• Regular users: Precomputed feed pushed to Redis
• Celebrities (>5K followers): Lazy fan-out at read time

Timeline = Precomputed_feed + Celebrity_tweets (merged at read)

Trending Topics

1. Count hashtag frequency in 1-hour sliding windows
2. Use Apache Flink for real-time stream processing
3. Store top-50 trending in Redis
4. Personalize by location

Search

• Elasticsearch indexes all tweets
• Inverted index on words
• Near real-time indexing via Kafka connector

Media Upload

Client → Upload Service → S3
                 ↓
          CDN Distribution

Conclusion

Twitter's architecture is a masterclass in read optimization — hybrid fan-out, Redis timelines, and Elasticsearch power the platform at scale.