Z2 · Scalability & Performance

Load balancer

Gold

A load balancer spreads incoming requests across several servers and routes around unhealthy ones, so capacity can grow by adding machines.

Use when: Two or more interchangeable servers · Rolling deploys and failover
Avoid when: A single server with no need for redundancy yet · Servers that keep per-user state, unless sessions move elsewhere
Trade-offs: One extra hop of latency · Layer 4 (fast, simple) vs layer 7 (routes by path or header) · Managed service cost vs running your own proxy
Requires: Server capacity & scaling up

3 objectives · 8 misconceptions

predict Predict per-server load when a load balancer fronts N servers.
explain Explain how health checks take failed servers out of rotation.
choose Choose between round-robin and least-connections when requests vary in cost.

Not true: Adding a load balancer by itself adds capacity.
Not true: The load balancer cannot be a single point of failure.
Not true: A load balancer fills one server before it uses the next.
Not true: A load balancer stops sending traffic to a dead server the moment it dies.
Not true: Sending every server the same number of requests balances the load.
Not true: Least-connections is always the better algorithm.
Not true: A load balancer's health checks fix or restart failed servers.
Not true: When one server behind a load balancer dies, the whole site goes down.

Learn

→Load balancer→