Load Balancer as a Service

The choice of algorithm directly affects performance. Round-robin is the simplest and is sufficient for servers of equal capacity and short requests. Least-connections is fairer for applications with variable request durations (ERP screens that generate reports, file uploads), because it does not send new load to a server already filled with slow requests. Weighted distribution is used when servers of different capacities are mixed. IP-hash and cookie-based persistence are necessary for legacy applications that keep session data in memory; however, in the long run, moving the session to a shared layer such as Redis and removing persistence is the right architecture for scaling flexibility, and we provide consulting on that transition as well.

The most insidious kind of outage is the "half-alive" server: the TCP port is open, but the application cannot connect to the database. A health check that only tests the port treats this server as healthy, and monitoring looks green while some of your users get errors. A correct setup is layered: a /health endpoint that verifies the application's own dependencies (database, cache, queue), plus an L7 check that queries this endpoint every 5 seconds, plus removal from the pool after 3 consecutive failures and gradual reinstatement (slow start). During setup we define this endpoint together with your development team; the real value of LBaaS shows in this fine-tuning.