About Load Balancing Basics

L4 vs L7, algorithms, health checks, failure modes. Part of the DevTools Surf developer suite. Browse more tools in the Info / Guides collection.

Use Cases

• DevOps engineers configuring Nginx or HAProxy for backend services
• Platform teams designing high-availability architectures with failover
• Backend developers understanding sticky sessions for stateful applications
• SREs planning capacity and choosing load balancing algorithms for traffic spikes

Tips

• Compare L4 vs L7 load balancing to pick the right layer for your needs
• Check the health check configuration to avoid routing to dead backends
• Review the failure modes section to plan for load balancer outages

Fun Facts

• Round-robin DNS was the earliest form of load balancing, used by NCSA in 1995 to distribute traffic for the first popular web server.
• L7 (application layer) load balancers can inspect HTTP headers, cookies, and URL paths — enabling features like sticky sessions and A/B testing at the routing level.
• The 'thundering herd' problem occurs when a load balancer's health check marks a backend as healthy and all queued requests rush to it simultaneously.

FAQ

Layer 4 vs Layer 7?

L4 balances on TCP/UDP (fast, protocol-agnostic). L7 balances on HTTP (can route by path, header, cookie). L7 is more common for web apps; L4 for low-latency or non-HTTP services.

Common algorithms?

Round-robin (simple), least-connections (balances load for long-lived connections), IP hash (sticky sessions). Weighted variants for heterogeneous backends.

What are health checks?

Periodic pings that remove unhealthy backends from rotation. Check something that exercises the service (not just TCP connect) to catch real failures.

Failure modes?

Cascading failures (retries slam a struggling service), thundering herd (all traffic goes to the healthy remaining instance), split brain (LB thinks service is down when it isn't).