Handling more load.
WHAT TO ESTABLISH FIRST
What the actual load is, and where the limit is.
WHY
Most scaling work addresses problems that do not exist.
WHAT TO MEASURE
Requests per second Response time distribution Resource utilisation Where time is spent
WHAT SCALING VERTICALLY MEANS
A larger machine.
WHAT IT PROVIDES
Simplicity, and it is frequently sufficient.
WHAT SCALING HORIZONTALLY MEANS
More machines behind a load balancer.
WHAT IT REQUIRES
Statelessness: no request depending on which instance handles it.
WHAT THAT FORBIDS
Session state in process memory Files written to local disk and read later
WHAT TO DO INSTEAD
Shared session storage, and shared file storage.
WHAT USUALLY BECOMES THE BOTTLENECK
The database.
WHAT TO DO ABOUT IT
Index properly Cache what is read repeatedly Use read replicas for read-heavy work Move heavy work to background processing
WHAT TO AVOID
Scaling before optimising Assuming the bottleneck rather than measuring it
