Load balancer

What it means

A load balancer is a device or service that spreads incoming requests across several servers so that none of them is overloaded. The user accesses a single website address and the load balancer decides which server will respond.

Besides distributing load, it regularly checks whether the servers are healthy. If one fails, it stops sending requests to it and redirects traffic to the others. This increases the availability of a website or application and allows individual servers to be maintained without downtime.

Requests can be distributed in different ways: in turn across all servers, by the current number of connections or so that one user stays on the same server for the whole session. In the cloud a load balancer is usually offered as a managed service that scales automatically.

How it works in practice

Load balancers are used for websites and applications where downtime means lost revenue or trust, such as online shops, booking systems or client portals. For a smaller company website, a single server and a CDN that spreads the load of static content are often enough.

Encrypted connections are often terminated at the load balancer, so this is where the SSL/TLS certificate for the whole service is managed. In a Kubernetes environment, an ingress component plays a similar role, routing requests to individual containers by host or path.

A load balancer also helps when deploying new versions. Traffic can be shifted gradually from the old version to the new one, and rolled back quickly if problems arise. When planning costs, include charges for the service itself and for data transferred.

What to watch out for

A load balancer on its own does not protect against a DDoS attack. It helps spread normal load, but a large attack can overwhelm it too. Protection requires combining the provider's defences, a CDN and traffic filtering rules.

If an application keeps login data only in one server's memory, a user moved to another server may be logged out. Before introducing a load balancer, check that the application is ready to run on several servers; otherwise hard-to-trace errors will appear.