[RFC]: Improved performance of enabling async_chunk in concurrent scenarios. E2E performance improved by 20% in a 10-concurrency scenario.

[amy-why-3459]

Motivation.

Improved performance of enabling async_chunk in concurrent scenarios. E2E performance improved by 20% in a 10-concurrency scenario.

Proposed Change.

Please provide the detailed design document of the RFC using the template.

Feedback Period.

No response

CC List.

No response

Any Other Things.

No response

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[Shirley125]

初版性能数据测试、输出初步blog；打点qwen3-omni高并发下耗时：异步等待存/取chunk排队时间较长
带上ziming修改4并发下async chunk e2e有收益，10并发下无收益

[Shirley125]

高并发下瓶颈：在队列中等待save chunk时间过长，recv_loop 空轮询次数过多，SHM get（反序列化）耗时偏高
在transfer adapter的recv_loop加指数退避sleep，提前过滤custom_process_func中返回None的任务，测试是否能减少cpu抢占