• BiscuitOS AI White Paper

  • LLM

    • Deepseek/Qwen

    • LLM Traning

    • LLM Fine-Tuning

    • LLM Quantization

    • LLM Inference

    • vLLM/Ollama/Pytorch

    • NVIDIA CUDA/PTX

  • AiOS

    • Load/Store/Atomic Roadmap

      • GPU Direct Peer ACCESS

      • CXL-MEMORY

      • HMM(Heterogeneous Memory Management): SVM/SPM

      • Nvidia-UVM

      • VRAM-MMIO/DEVICE-ZONE

    • MEMORY READ/WRITE Roadmap

      • DMA/SGDMA/Coherent-DMA/Streaming-DMA/CMA

      • P2PDMA

      • GPU P2PDMA

      • IOMMU

      • VFIO

      • DMA-BUF

      • DMA-POOL

    • Pooling Roadmap

      • RDMA-OpenShmem

      • NvSHMEM

      • CXL Pooling/CXL-Sharing

    • Heterogeneous Other Roadmap

      • DAX/FSDAX/PMEM

      • CXL 1.0/2.0

      • MMIO/PIO

      • TIERED-MEMORY

      • NUMA

  • AI Infra

    • C2C Interconnect

      • Intel QPI/UPI/MESH

      • AMD Infinity Fabric

    • Heterogeneous Interconnect

    • Hardware

      • CPU(Intel/AMD/ARM)

      • GPU(Nvidia/AMD)

      • CXL/HBM/DDR

      • NIC

  • PCI(Peripheral Component Interconnect)

  • PCIe(Peripheral Component Interconnect Express)

  • CXL(Compute Express Link)

  • HMM(Heterogeneous Memory Management)

  • I/O Space

  • MMIO(Memory-Mapped I/O)

  • DMA(Direct Memory Access)

    • Coherent DMA

    • Streaming DMA

    • DMA-BUF

    • DMA Pool

    • CMA

    • SWIOTLB

  • RDMA(Remote Direct Memory Access)

  • SGDMA(Scatter-Gather Direct Memory Access)

  • IOMMU(Input-Output Memory Management Unit)

  • VFIO(Virtual Function I/O)

  • P2PDMA(Peer-to-Peer Direct Memory Access)

  • NUMA(Non-Uniform Memory Access)

    • LOCAL MEMORY

    • REMOTE MEMORY

    • NUMA BALANCING

  • HBM(High Bandwidth Memory)

  • DEVICE MEMORY

    • DAX(Direct Access)

    • FSDAX(File System Direct Access)

    • DIRECT-IO

    • BUFFER-IO

    • DEVICE ZONE

    • HBM/GDDR

    • VRAM

  • TIERED MEMORY Technology

  • NVDIMM/PMEM(Persistent Memory)

  • NPU/xPU

  • General Filesytem Protocl

    • POSIX Interface Layer

    • SystemCall Layer

    • Virtual Filesytem Layer(VFS)

    • Filesytem Implementation layer

      • EXT2/EXT3/EXT4

      • XFS

    • Page-CACHE Layer

    • Block I/O Layer(BIO)

    • Storage Device Driver

  • High-Performance Storage

    • FUSE

    • SPDK

    • NVMe/NVMe-of

    • IO-Uring

    • Mooncake.STORE L3/L3.5 Layer

    • GPUDirect Storage(GDS)

  • Extension Technologies

    • DAX

    • FSDAX

    • DirectIO

    • Buffer-IO

    • File-Mapping Mechanism

    • FADVISE(File Advise)

    • FS-ZEROCOPY(splice/vmsplice/sendfile)

    • HugeTLBFS

    • Fallocate

    • MEMFD

    • READAHEAD

    • inotify/fanotify

  • Storage Hardware

    • SSD

    • HDD

    • NVIDIMM(PMEM)

    • Zoned SSD

  • Linux Kenrel Network Protocol Stack

    • Socket Layer

    • Transport Layer: TCP/UDP/SCTP/DCCP

    • Network Layer: IP

    • Data Link Layer

    • Network Device Driver

  • High-Performance Networking

    • DPDK

    • RDMA

    • NTB

    • FC

    • GPUDirect RDMA(GDR)

  • Extension Technologies

  • Networking Hardware

    • Standard Ethernet TCP/IP Fabric

    • InfiniBand(IB) and RoCEv2 Networks

    • Fibre Channel(FC) Network

  • ZERO COPY Optimization

  • NUMA BLANCING/SCHEDULE Optimization

  • PARALLEL Optimization

  • Algorithm Visualization

  • Classic Kernel Data Structures

    • Classic Linked list

    • Red-Block Tree

    • Xarray

    • Radix Tree

    • Hash list

    • Hash Table

    • Interval Tree

  • Performance Analysis Tools