When traffic spikes overwhelm a system, standard defenses fail. Engineers must turn to Load Shedding, a strategy that deliberately drops requests to save the whole system. This prevents a total System Collapse where every user gets a slow, error-filled experience. By sacrificing a portion of traffic immediately, you preserve capacity for everyone else.

Unlike rate limiting, which targets specific abusive users, load shedding is self-preservation. It’s the server admitting it is overwhelmed and dropping traffic regardless of the source. Companies like Amazon and Netflix practice this via Graceful Degradation, turning off non-essential features like recommendations to keep core functions like checkout or video playback running smoothly.

The key is prioritizing requests by criticality. A payment transaction is vital, while an analytics ping is not. Modern systems monitor concurrent requests, sending a fast HTTP 503 error to number 501 if the limit is 500. This 'fast failure' preserves CPU cycles, preventing the server from wasting resources on requests the user won't receive.