As load increases, new servers spin up automatically. When load drops, they terminate. Watch it happen below — servers spawn under load, die when idle.
INCOMING TRAFFIC LOAD
LOWMEDIUMHIGHPEAK
SERVER POOL
Server 1Always on
Server 2Always on
Server 3Auto spawn
Server 4Auto spawn
Server 5Auto spawn
Fading servers = auto-scaling in/out based on load
● LIVE
VERTICAL SCALE ↑
🖥
Bigger machine. More RAM, more CPU. Has a physical limit. Single point of failure.
HORIZONTAL SCALE →
🖥🖥🖥
More machines. Infinite theoretical limit. Fault tolerant. The cloud way.
When you add/remove servers, consistent hashing ensures only K/N keys are remapped (not everything). Used by Cassandra, DynamoDB, Akamai CDN to minimize cache invalidation on cluster resize.