Feature | Description |
---|---|
Behavior | Requests rotate in order (Target A → B → C → A…) |
Best For | Uniform backends with similar capacity |
Supports Weights | ✔ Yes |
Example Use Case | Balanced LLM provider pool with similar models |
Feature | Description |
---|---|
Behavior | Targets with higher weight receive more requests |
Best For | Backends with unequal capacity or cost |
Customization | ✔ Fine-grained traffic control |
Example Use Case | 60% traffic to OpenAI, 40% to Anthropic |
Feature | Description |
---|---|
Behavior | Tracks real-time active connections |
Best For | Long-lived or uneven request durations |
Adaptability | ✔ Dynamically shifts based on load |
Example Use Case | Chat or streaming-based APIs |
Feature | Description |
---|---|
Behavior | Randomized request routing |
Best For | Simple, stateless services |
Overhead | ✔ Minimal tracking required |
Example Use Case | Lightweight APIs where simplicity matters |
Scenario | Recommended Algorithm |
---|---|
Uniform backend capacity | Round-Robin |
Different backend resources | Weighted Round-Robin |
Long-lived connections | Least Connections |
Stateless quick responses | Random |