Scaling for Millions of Users

Scenario : A system that caters large number of users with a web front-end and a mobile app.

Things to consider

Choice of Database
- For most cases relational database like MySQL would work
- Things to consider when going for NoSQL
  - Application requires super law latency
  - Unstructured data or data has no relations
  - Need to store a massive amount of data
Scale the application tier
- Decide whether horizontal or vertical scaling
- Vertical scaling is limited by CPU, Memory, Space and cost
- Horizontal scaling require adding new components like load balancers and changing the application to be able to share load
Scale the database (database-scaling)
- Database Replication
- Database Sharding
Use Caching and CDN for static assets
- Avoid frequent calls to database
- Improve performance
Make the application/web tier stateless
- Mitigates issue with stateless web tier/sticky sessions
- Easier horizontal scaling
Support multiple data centers
- Geo Routing
- Geo DNS
- Disaster Recovery
- Availability
Decouple the components and make processes async, Introduce message queues
- RabbitMQ
- Kafka

References: