DingoFS is a cloud-native distributed high-speed file storage system designed and developed by DataCanvas. It integrates multiple features such as elasticity, multi-cloud compatibility, multi-protocol convergence, and exceptional performance.By leveraging its multi-tiered, multi-type, and high-performance distributed multi-level caching architecture, DingoFS accelerates data I/O for AI workflows, effectively addressing burst I/O challenges in AI scenarios. Additionally, it provides local cache storage capabilities to meet the full lifecycle storage requirements of large-scale AI models.
1. 📂 POSIX Compliance
DingoFS delivers a native file system-like operational experience, enabling seamless system integration.
2. 🤖 AI-Native Architecture
Deeply optimized for large language model (LLM) workflows, efficiently managing massive training datasets and checkpoint workloads.
3. ☁️ S3 Protocol Compatibility
DingoFS supports standard S3 interface protocols for streamlined access to filesystem namespace resources.
4. 🔗 Fully Distributed Architecture
DingoFS's metadata Service (MDS), data storage layer, caching system, and client components all support linear scalability.
5. ⚡ Exceptional Performance
Combines SSD-level low-latency responsiveness with object storage-grade elastic throughput capacity.
6. 🚀 Intelligent Caching Acceleration System
DingFS implements a three-tier caching topology (memory/local SSD/distributed cluster) to deliver high-throughput, low-latency intelligent I/O acceleration for AI workloads.
All Documentation Docs
| Release | Version |
|---|---|
| Debian | 9+ |
| Centos | 7+ |
| Ubuntu | 20+ |
Deploying a cluster via DingoAdm requires the following two network prerequisites:
-
The control machine where DingoAdm is installed must have SSH access to the target servers for service deployment.
-
All servers must be able to pull images from the Docker registry.
We recommend Rocky and Ubuntu to develop the DingoFS codebase.
sudo dnf install -y epel-release
sudo dnf install -y gcc-toolset-13* libnl3-devel libunwind-devel python3-devel
wget http://github.com/Kitware/CMake/releases/download/v3.30.1/cmake-3.30.1-linux-x86_64.tar.gz
tar zxvf cmake-3.30.1-linux-x86_64.tar.gz
sudo cp -rf cmake-3.30.1-linux-x86_64/bin/* /usr/local/bin/ && sudo cp -rf cmake-3.30.1-linux-x86_64/share/* /usr/local/share && rm -rf cmake-3.30.1-linux-x86_64
source /opt/rh/gcc-toolset-13/enablesudo apt update
sudo apt install -y make gcc g++ libnl-genl-3-dev libunwind-dev python3-dev
wget http://github.com/Kitware/CMake/releases/download/v3.30.1/cmake-3.30.1-linux-x86_64.tar.gz
tar zxvf cmake-3.30.1-linux-x86_64.tar.gz
sudo cp -rf cmake-3.30.1-linux-x86_64/bin/* /usr/local/bin/ && sudo cp -rf cmake-3.30.1-linux-x86_64/share/* /usr/local/share && rm -rf cmake-3.30.1-linux-x86_64We recommend using GCC 13 as the primary compiled language.
- Dingo-eureka: A Necessary Service Components for DingoFS.
- Dingo-sdk: A Unified Software Development Kit (SDK) required for DingoFS.
wget http://github.com/jemalloc/jemalloc/releases/download/5.3.0/jemalloc-5.3.0.tar.bz2
tar -xjvf jemalloc-5.3.0.tar.bz2
cd jemalloc-5.3.0 && ./configure && make && make installIf you installed the software using a Docker container, the container already includes pre-integrated Dingo-eureka and Dingo-sdk.
git submodule sync
git submodule update --init --recursivemkdir build
cd build
cmake ..
make -j 32If you want use 3FS as your block cache, you should build with WITH_LIBUSRBIO option:
cmake -DWITH_LIBUSRBIO=ON ..
DingoFS is Sponsored by DataCanvas, a new platform to do data science and data process in real-time.
DingoFS is an open-source project licensed under the Apache License Version 2.0, welcome any feedback from the community. For any support or suggestion, please contact us.
If you have any technical questions or business needs, please contact us.
Attach the Wetchat QR Code
