Mastering Load Balancing: Types, Architectures, and Algorithms Explained

Classification and Architecture

Common load balancers are divided into three types: DNS load, hardware load, software load.

1. DNS Load Balancing

Definition: Resolving the same domain name to different IP addresses, typically used for geographic load balancing. For example, northern and southern users receive different addresses.

Advantages: Simple, low cost; enables nearby access, improving speed.

Disadvantages: Updates are not timely due to DNS cache; poor scalability; simple allocation strategies.

2. Hardware Load Balancing

Definition: Implemented via dedicated hardware devices (e.g., F5, A10) that perform load balancing.

Advantages: Powerful features (supports multiple layers, many algorithms, global load balancing); high performance (over 1 million concurrent connections); high stability; includes security functions (firewall, DDoS protection).

Disadvantages: Expensive (starting around 150,000 CNY, up to millions); limited extensibility.

3. Software Load Balancing

Definition: Implemented with load‑balancing software such as Nginx or LVS.

Advantages: Simple deployment and maintenance; inexpensive (just a Linux server); flexible and customizable.

Disadvantages (compared with hardware): Moderate performance (Nginx handles about 50,000 concurrent connections); fewer features; generally lacks firewall and DDoS protection.

4. Typical Architecture

The three mechanisms can be combined. Basic principle: DNS load balancing for geographic level; hardware load balancing for cluster level; software load balancing for machine level.

Algorithms

Based on the goal, algorithms are classified into four categories: task classification (even distribution by count, proportion, weight), load‑balancing (based on server load metrics like CPU, connections, I/O), performance‑optimal (assign to servers with fastest response), and hash‑based (assign based on hash of key information).

1. Round‑Robin

Definition: Assign tasks to servers in order, without considering server status. Simple but may ignore server health.

2. Weighted Round‑Robin

Definition: A variant of round‑robin that uses static weights based on hardware capacity to distribute tasks. Still does not consider real‑time server status.

3. Least‑Load First

Definition: Assign tasks to the server with the lowest current load, measured by metrics such as connections, request count, CPU, or I/O. Improves over round‑robin but adds complexity due to state monitoring.

4. Performance‑Optimal

Definition: From the client’s perspective, assign tasks to the server with the fastest processing speed. Similar to least‑load but requires extensive monitoring and analysis.

5. Hash‑Based

Definition: Compute a hash on certain key information of the task; requests with the same hash value are routed to the same server, useful for session affinity or transactional workloads (e.g., source‑IP hash, ID hash).