Architecting a large social network without breaking the bank
summary
Goal is 50 million users with heavy engagement and handling up to 5,000 writes/sec with 99.99% of writes under 200ms by using a queue-first ingest (accept, ack, background work), a CDN/S3/R2 for media, a distributed key-value store and queues for storage and smoothing spikes. Design forbids edits (no edits delete wins), uses batching for timelines, treats whales by client-side merge, and replaces Snowflake with a 16-byte time‑machine‑id to avoid predictable ID scanning.
AI-written summary · about
chapters (8)
- 0:00Introduction and philosophy
- 4:09Analyzing social network load
- 11:38Design requirements
- 20:45The architecture strategy
- 29:06Data handling and ID logic
- 37:13Data distribution patterns
- 46:43Timeline and scaling logic
- 55:17System summary and closing
related talks
17mExtreme Engineering-Elements of Building a High -scale Broking Application
56mMastering Cloud Scalability and System Designs Deep dive
46mDigital scale-up architectural patterns
41mDesign (Failures) Case Study
43mArchitecting for Speed & Scale
01h 14mRead and write considered harmful
40mArchitecting for Data-Intensive Workloads
01h 06mThe Legends of Twitter and Beyond: Real-World Architectures at Scale Panel