The DMA Streaming Framework: Kernel-Level Buffer Orchestration for High-Performance AI Data Paths

Researchers have developed dmaplane, a Linux kernel module that provides buffer orchestration for AI workloads, addressing the gap between efficient data transport and proper buffer management. The system integrates RDMA, GPU memory management, and NUMA-aware allocation to optimize high-performance AI data paths at the kernel level.

• →dmaplane introduces kernel-level buffer orchestration as a missing layer between AI transport libraries and hardware.
• →The system provides a stable kernel API through /dev/dmaplane with ring-based command channels and DMA buffer lifecycle management.
• →Integration includes GPU memory support via PCIe BAR pinning and cross-device sharing through dma-buf export.
• →Researchers demonstrated end-to-end disaggregated inference by transferring KV-cache chunks between machines using RDMA.
• →The framework addresses NUMA cross-node penalties and provides credit-based flow control for sustained RDMA workloads.