Horizontal Scaling

Horizontal scaling (also called scaling out) increases capacity by adding more machines or instances to handle load, distributing traffic across them via a load balancer. The opposite is vertical scaling (or scaling up), which adds CPU, RAM, or disk to existing machines.

Horizontal scaling has two practical advantages over vertical scaling: there's no upper bound (you can always add another server), and the system becomes fault-tolerant by default (if one instance fails, traffic routes to the others). The trade-off is that horizontal scaling requires stateless application design — sessions, file uploads, or in-memory caches that exist on only one instance break the moment a request lands on a different node. State has to move to a shared store: a database for sessions, object storage for files, Redis for caches.

Real-world horizontal scaling almost always combines several patterns: stateless application servers behind a load balancer, a managed database with read replicas, a distributed cache layer, and either container orchestration (Kubernetes) or a PaaS auto-scaler that adds and removes instances based on CPU or request-rate thresholds. The biggest gotcha is sticky sessions — they sidestep statelessness in the short term but undermine fault tolerance and force uneven load distribution.