A load balancer sits in front of a fleet of backend instances and spreads incoming requests across them — by round robin, least-connections, or a hashing scheme — so traffic is shared rather than piling onto one instance. Paired with health checks, it also routes around instances that stop responding, turning a fleet of individually fragile machines into a system that tolerates individual failures.
Worked example: a reverse proxy (nginx, ALB, Envoy) spreads incoming requests across a pool of backend instances so none saturates; algorithms include round-robin, least-connections, and consistent-hash (for cache affinity). Gotcha: naive round-robin ignores that requests cost different amounts, so one slow endpoint piles up on a few instances; and the balancer itself is a single point of failure unless made redundant (multiple LBs behind DNS or anycast).