Sign inSign up

openeuler/mooncake

Sponsored OSS

By openeuler

•Updated 10 days ago

Image
0

4.2K

openeuler/mooncake repository overview

⁠Quick reference

⁠Mooncake | openEuler

Current Mooncake docker images are built on the openEuler⁠. This repository is free to use and exempted from per-user rate limits.

Mooncake is a high-performance transfer engine for distributed AI inference workloads, developed by KVCache.AI. It provides a RDMA/TCP-based data plane for efficient KV cache transfer between disaggregated prefill and decode instances in large language model (LLM) serving systems. Mooncake is designed for distributed storage and high-speed data movement in cloud-native AI infrastructure.

Learn more about Mooncake on GitHub⁠⁠.

The tag of each mooncake docker image is consist of the version of mooncake and the version of basic image. The details are as follows

TagCurrentlyArchitectures
0.3.13-oe2403sp4⁠mooncake 0.3.13 on openEuler 24.03-LTS-SP4amd64, arm64
0.3.11.post1-oe2403sp4⁠Mooncake 0.3.11.post1 on openEuler 24.03-LTS-SP4amd64, arm64
0.3.11.post1-oe2403sp3⁠Mooncake 0.3.11.post1 on openEuler 24.03-LTS-SP3amd64, arm64

⁠Usage

In this usage, users can select the corresponding {Tag} based on their requirements.

  • Pull the openeuler/mooncake image from docker

    docker pull openeuler/mooncake:{Tag}
    
  • Run the transfer engine benchmark (sender side)

    docker run --rm --network host openeuler/mooncake:{Tag} \
        transfer_engine_bench --mode=initiator \
        --metadata_server=<metadata-ip>:2379 \
        --local_server_name=<local-ip>:12345
    
  • Run the transfer engine benchmark (receiver side)

    docker run --rm --network host openeuler/mooncake:{Tag} \
        transfer_engine_bench --mode=target \
        --metadata_server=<metadata-ip>:2379 \
        --local_server_name=<local-ip>:12346
    
  • Validate the transfer engine

    docker run --rm --network host openeuler/mooncake:{Tag} \
        transfer_engine_validator --metadata_server=<metadata-ip>:2379
    
  • View available tools

    ToolDescription
    transfer_engine_benchBenchmark the transfer engine throughput and latency
    transfer_engine_validatorValidate transfer correctness between two nodes
    transfer_engine_bench_with_notifyBenchmark with notification-based completion

⁠Question and answering

If you have any questions or want to use some special features, please submit an issue or a pull request on openeuler-docker-images⁠.

Tag summary

Content type

Image

Digest

sha256:06bdb2ea5…

Size

367.2 MB

Last updated

10 days ago

docker pull openeuler/mooncake

This week's pulls

Pulls:

35

Last week