MooncakeConnector Usage Guide

About Mooncake

Mooncake aims to enhance the inference efficiency of large language models (LLMs), especially in slow object storage environments, by constructing a multi-level caching pool on high-speed interconnected DRAM/SSD resources. Compared to traditional caching systems, Mooncake utilizes (GPUDirect) RDMA technology to transfer data directly in a zero-copy manner, while maximizing the use of multi-NIC resources on a single machine.

For more details about Mooncake, please refer to Mooncake project and Mooncake documents.

Prerequisites

Installation

Install mooncake through pip: uv pip install mooncake-transfer-engine-cuda13.

vLLM defaults to CUDA 13. On a CUDA 12 environment install mooncake-transfer-engine instead — the two are the same release built against different CUDA majors, and the wrong one fails to import with libcudart.so.<major>: cannot open shared object file.

Refer to Mooncake official repository for more installation instructions

Usage

Prefiller Node (192.168.0.2)

bash

vllm serve Qwen/Qwen2.5-7B-Instruct --port 8010 --kv-transfer-config '{"kv_connector":"MooncakeConnector","kv_role":"kv_producer"}'

Decoder Node (192.168.0.3)

bash

vllm serve Qwen/Qwen2.5-7B-Instruct --port 8020 --kv-transfer-config '{"kv_connector":"MooncakeConnector","kv_role":"kv_consumer"}'

Proxy

bash

python examples/disaggregated/mooncake_connector/mooncake_connector_proxy.py --prefill http://192.168.0.2:8010 --decode http://192.168.0.3:8020

Now you can send requests to the proxy server through port 8000.

Environment Variables

VLLM_MOONCAKE_BOOTSTRAP_PORT: Port for Mooncake bootstrap server
- Default: 8998
- Required only for prefiller instances
- For headless instances, must be the same as the master instance
- Each instance needs a unique port on its host; using the same port number across different hosts is fine
WITH_NVIDIA_PEERMEM: Selects how mooncake registers GPU memory for RDMA. Read by mooncake, not vLLM.
- Default: 1, which uses ibv_reg_mr() and requires the nvidia-peermem kernel module to be loaded
- Set to 0 to use the DMA-BUF path, which does not need that module. Required on hosts where nvidia-peermem is not loaded, such as GB200
- With the container image, pass it at run time: docker run -e WITH_NVIDIA_PEERMEM=0 ...
- Symptom when left unset on such a host: Failed to register memory <addr>: Bad address [14] from rdma_context.cpp, and KV transfers fail
VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT: Timeout (in seconds) for automatically releasing the prefiller’s KV cache for a particular request. (Optional)
- Default: 480
- If a request is aborted and the decoder has not yet notified the prefiller, the prefill instance will release its KV-cache blocks after this timeout to avoid holding them indefinitely.

KV Transfer Config

KV Role Options

kv_producer: For prefiller instances that generate KV caches
kv_consumer: For decoder instances that consume KV caches from prefiller
kv_both: Enables symmetric functionality where the connector can act as both producer and consumer. This provides flexibility for experimental setups and scenarios where the role distinction is not predetermined.

kv_connector_extra_config

num_workers: Size of thread pool for one prefiller worker to transfer KV caches by mooncake. (default 10)
mooncake_protocol: Mooncake connector protocol. (default "rdma")
device_name: Comma-separated whitelist of RDMA devices (e.g. "mlx5_0,mlx5_1") to restrict topology discovery to. Empty discovers every device. Useful on hosts exposing a mix of InfiniBand and RoCE ports, where both peers must settle on the same link layer.

Example Scripts/Code

Refer to these example scripts in the vLLM repository: