How LVS Chooses Servers: Exploring Round‑Robin, Weighted and Least‑Connection Algorithms

LVS (Linux Virtual Server) decides which real server will handle a client request using various scheduling algorithms. Below are the most commonly used methods.

1. Round‑Robin

The simplest algorithm cycles through the servers in the cluster, assigning each new request to the next server regardless of its current load. It works best when all servers have similar performance.

i = -1;
i = (i + 1) mod n;

2. Weighted Round‑Robin

Each server is assigned a weight reflecting its processing capacity. Requests are still distributed in a round‑robin fashion, but a server is selected only when its weight permits, allowing more powerful servers to receive more traffic.

For example, with three servers weighted 4, 3, 2 and nine incoming requests, the distribution follows the weighted sequence.

Below is a JavaScript simulation of the algorithm:

// JavaScript example (simplified)
var weights = [4,3,2];
var servers = ['A','B','C'];
var i = -1;
for (var req=0; req<9; req++) {
  i = (i + 1) % servers.length;
  // decide based on weight...
}

3. Least Connections

This dynamic algorithm sends a request to the server that currently has the fewest active connections. It is similar to round‑robin but better suited when server capacities differ.

The scheduler tracks each server’s connection count, selects the minimum, increments the count for the chosen server, and decrements it when the connection ends or times out.

4. Weighted Least Connections

Combines the concepts of weight and connection count. For each server, the value connections / weight is computed; the server with the smallest result receives the next request, balancing load according to both capacity and current usage.