Inquiry Regarding Prefix Caching in the vllm Baseline

[Lin-Qingyang-Alec]

I hope this message finds you well. I recently read your technical report, and I found it very insightful. Thank you for sharing your work!

While reviewing the experimental details, I noticed some aspects related to the baseline were not entirely clear. Specifically, you mentioned using the vllm service as a baseline for Mooncake. I observed that Mooncake utilizes prefix caching technology, and I noticed that vllm has a startup parameter: --enable-prefix-caching, which seems to serve a similar purpose.

Could you kindly clarify whether you enabled this feature in the baseline during your experiments?

Thank you for your time, and I appreciate any insights you can provide.

Best regards.

[james0zan]

The support of prefix cache in vLLM is later than our experiment thus we do not compare with it.