Dispatch Policy

## Overview

The Dispatch Policy determines how the system allocates rendering servers to clients. Two allocation modes are available: **General Mode** and **Simple Mode**.

![](https://docs.paraverse.cc/media/202504/cc7f9a1470ff4ab08397c3091fd95a156611.png)

## Allocation Priority (General Mode)

The general dispatch follows this priority order:

**Specify Group > Specify Region > Closest Distance > Pre-started Instances > Minimum Concurrency > Synchronized Applications > Low Resource Utilization**

### Detailed Rules

- **Group:** If a group is specified, allocation is restricted to that group only.
- **Region:** If a region is specified, the system prioritizes allocation within that region. If resources are insufficient, it falls back to other regions.
- **No Group/Region:** Allocation defaults to the server cluster partition closest to the client. If that partition has insufficient resources, the system allocates randomly to other regions.
- **Pre-started Instances:** If a node has available pre-started instances, the client is allocated directly to that node.
- **Concurrency:** Nodes are filtered by concurrency load; lower-concurrency nodes are preferred.
- **App Sync:** The system checks application synchronization status and prioritizes nodes where the application is already synchronized.
- **Resource Usage:** Allocation is halted for any node where **GPU, CPU, or memory usage exceeds 85%**.

## Simple Mode

For high-concurrency scenarios requiring optimization, a **Simple Mode** is available. When selected, the following applies:

- Best suited for **public cloud environments** where applications are packaged as images and servers are created in bulk.
- The system **skips synchronization record checks** and server heterogeneity detection.
- All servers are treated as having synchronized applications by default, and all servers are assigned **equal weight** during allocation.

This mode reduces overhead and improves throughput in scenarios where server homogeneity can be assumed.