Dispatch Policy

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

**Simple Mode Allocation Strategy Considerations:**

In the case of system optimization for high concurrency, a "simple" allocation strategy has been added. The considerations when selecting the simple mode are as follows:

1. Generally applicable in public cloud environments where applications are packaged in images and used for scenarios involving the bulk creation of servers.

2. In the simple mode, the system no longer detects synchronous records and server heterogeneity factors. Each server is assumed to have synchronized applications by default, and all servers have equal weight.

**General Allocation Strategy Rules:**

- Specify group > Specify region > Prefer closest distance > Prefer existing pre-started instances > Prefer minimum concurrency > Prefer synchronized applications > Prefer low resource utilization

**Specific rules include:**

- If a group is specified, allocation is limited to that group only.

- If a region is specified, allocation is prioritized in that region. If resources in that region are insufficient, allocation can be made to other regions.

- If no group or region is specified, allocation is based on the client's nearest server cluster partition. If resources in that partition are insufficient, allocation will be made randomly to other regions.

- If the current node has spare pre-started instances, allocation is direct to that node.

- Nodes are filtered by concurrency, with allocation prioritized to nodes with lower concurrency.

- During allocation, application synchronization is checked, and allocation is prioritized to nodes with synchronized applications.

- Resource usage is monitored, and allocation will not continue if any GPU, CPU, or memory usage exceeds 85%.

By following these rules, the system can effectively allocate servers in high concurrency scenarios, improving system performance and resource utilization.