Why placement matters
A load balancer spreads traffic across many servers so no single node is overwhelmed. Where you put it decides what it can balance.
- At the edge it sits in front of web servers and distributes public requests.
- Internal balancers sit between tiers, spreading calls from app servers to backend services.
- A single balancer is itself a single point of failure, so it usually runs as a redundant pair.
Common topology
Traffic enters one public balancer, fans out to a pool of stateless web servers, and internal balancers route onward to services.
Health and failover
- Balancers run health checks and stop sending traffic to unhealthy nodes.
- Use multiple balancers across zones so one failure does not take down entry.
- DNS or an anycast address fronts the balancer pair for redundancy.
Algorithms
- Round robin spreads evenly when nodes are equal.
- Least connections favors the least busy node.
- Hashing can route the same client to the same node when needed.
Key idea
Place load balancers at every fan out point and make them redundant, since a balancer that cannot fail over is just a bottleneck.