How Load Balancers Route Millions of Requests Per Second Across Servers

A single DNS name can absorb tens of millions of requests per second, and the load balancer is the largely invisible component that makes this possible. Its job is to choose, within microseconds, which of many identical servers should handle each request while adapting to crashes, traffic spikes, and uneven workloads. Simple round-robin distribution can overload servers when some requests are much heavier than others, so smarter methods such as least connections and the power-of-two-choices approach are used to balance work more effectively. Modern systems stack several layers of balancing, from anycast routing that sends users to the nearest data center, to transport-level routing based on connection details, to application-level routing that reads request contents. Together these layers distribute traffic and absorb attacks across a wide footprint rather than concentrating them on a single target.