Course contentsShow
← System Design
Lesson 86 of 1,9193. Load BalancingFree lesson

Weighted Round Robin

Extends round robin by assigning weights to servers based on capacity, sending more traffic to powerful servers.

Weighted Round Robin

What you'll learn: How load balancers can send more traffic to powerful servers using weighted assignments.


The Core Idea

Basic round robin treats all servers equally, rotating requests one at a time. But what if your servers aren't equal? Perhaps you have three servers: one beefy machine with 16 cores and two smaller ones with 4 cores each. Sending them equal traffic wastes the big server's capacity.

Weighted Round Robin solves this by assigning each server a weight that reflects its capacity. The load balancer then distributes requests proportionally to these weights.

How It Works

Instead of Server1 → Server2 → Server3 → repeat, you assign weights:

  • Server 1 (powerful): weight = 4
  • Server 2 (moderate): weight = 2
  • Server 3 (moderate): weight = 2

Now the pattern becomes: S1 → S1 → S2 → S1 → S3 → S1 → S2 → S3 → repeat

Server 1 receives 4 out of every 8 requests (50%), while Servers 2 and 3 each get 2 requests (25% each).

When to Use It

This algorithm shines when you have heterogeneous servers—machines with different CPU, memory, or network capabilities. It's also useful during:

  • Rolling upgrades: Give new servers lower weights initially
  • Cost optimization: Mix powerful and modest instances
  • Geographic proximity: Weight nearby servers higher for latency

The load balancer administrator configures weights based on capacity measurements or estimated throughput limits.

Limitations

Weighted Round Robin still doesn't consider real-time server health or current load—it mechanically follows the weight pattern regardless of whether Server 1 is actually overloaded at that moment.

Key Takeaway: Weighted Round Robin extends basic round robin by assigning capacity-based weights to servers, directing proportionally more traffic to powerful machines while still maintaining predictable rotation.