수평 확장과 수직 확장 비교

When an API runs out of capacity you have two levers:

• Vertical scaling (scale up) — make the machine bigger
• Horizontal scaling (scale out) — add more machines

Each has very different cost and reliability profiles.

Vertical scaling means upgrading a single server: more CPU cores, more RAM, faster disks.

It is simple — no code changes — but you eventually hit the largest instance available, and that one box is a single point of failure.

Horizontal scaling adds more identical servers behind a load balancer. Traffic spreads across the fleet.

• No hard ceiling — add nodes as needed
• One node failing does not take down the service

Vertical scaling cost rises steeply — top-tier hardware carries a premium. Horizontal scaling uses many commodity nodes, which is usually cheaper per unit of capacity at large scale.

To scale out, any node must handle any request. That requires stateless services — no session data stored on the instance.

Push session and state to a shared store (Redis, a database) so nodes stay interchangeable.

Cloud platforms add or remove nodes automatically based on metrics like CPU or request rate.

You define a min, max, and a target metric; the platform keeps the fleet sized to load.

Scaling up is the right call when:

• The workload is hard to distribute (a single large in-memory dataset)
• You need a quick fix before re-architecting
• Licensing is per-node and a bigger box is cheaper

Adding nodes is not free scaling — shared resources (a single database, a lock) become the new bottleneck. This is why scaling the data tier often matters more than the app tier.

Real systems mix strategies: right-size each node (a bit of vertical) then run many of them (horizontal). The goal is the best cost per request at your reliability target.

Never scale blindly. Profile first to find the real constraint — CPU, memory, I/O, or a downstream dependency. Scaling the wrong dimension wastes money and hides the true bottleneck.

Capacity is not free. A fleet sized for peak sits idle at night, burning money. Combine auto-scaling with right-sizing and consider spot or reserved capacity to match spend to real demand.

You compared scaling strategies:

• Vertical — bigger box, simple, but capped and a single point of failure
• Horizontal — more boxes, resilient, needs statelessness
• Auto-scaling sizes the fleet to load
• Always measure the real bottleneck first