Phoenix Documentation
Welcome to the Phoenix documentation. Phoenix is a middleware that enables direct I/O transfer from storage systems to xPU accelerators (GPU/NPU), bypassing CPU bounce buffers (“phony buffers”). It exposes a kernel module, a user-space library, and pybind-based adapters for AI applications (e.g. vLLM model loading).
Document index
Document |
Description |
|---|---|
System architecture, data path, and directory responsibilities |
|
Prerequisites, build, kernel-module install, and quick demo |
|
|
|
User-space C/C++ library API ( |
|
Application adapters: vLLM (available) and lmcache (integration guide) |
|
GPU vendor porting guide (kernel P2P backend + user-space connector), bilingual 中英双语 |
|
Kernel-module install troubleshooting |
|
Engineering tasks and research directions |
Quick links
Project README:
README.mdContributor guide:
CONTRIBUTING.mdAI developer orientation:
AGENTS.md