How Nginx’s Smooth Weighted Round‑Robin Load Balancing Works

Weighted Round‑Robin in Nginx

Nginx supports five load‑balancing algorithms; this article concentrates on the weight‑based round‑robin method, where each server is assigned a weight value and the probability of being selected is proportional to that weight.

upstream backend {
    server 10.xxx.xxx.1 weight=10;
    server 10.xxx.xxx.2 weight=20;
}

After a full cycle, the server with weight 10 will be chosen roughly half as often as the server with weight 20 (a 1:2 ratio). The basic round‑robin simply picks the next node in order, e.g., {a, b, c}, which can lead to bursts when weights differ.

When weights differ, the naive weighted round‑robin may produce a sequence such as {a, a, a, a, b, b, c}, concentrating load on the high‑weight server.

Smooth Weighted Round‑Robin

Algorithm is as follows: on each peer selection we increase current_weight of each eligible peer by its weight, select the peer with the greatest current_weight and reduce its current_weight by the total number of weight points distributed among peers.

The algorithm executes in two steps for each selection:

Increase each node’s current_weight by its own weight.

Select the node with the highest current_weight and subtract the sum of all weights from that node’s current_weight.

For example, with nodes a, b, c having weights 5, 1, 1, the selection sequence becomes {a, a, b, a, c, a, a}, which is smoother than the naive sequence {c, b, a, a, a, a, a}.

The proof shows that after each full round the sum of all current_weight values returns to zero, guaranteeing that each node is chosen a number of times proportional to its weight and that no node is selected repeatedly in a burst.

Conclusion

The smooth weighted round‑robin algorithm can be implemented with just two simple steps and provides fair, weight‑proportional distribution of requests across backend servers.