The DMA Streaming Framework: Kernel-Level Buffer Orchestration for High-Performance AI Data Paths
This work addresses performance bottlenecks in AI systems for developers and researchers, though it is incremental as it builds on existing transport libraries by adding a kernel-level orchestration layer.
The paper tackles the problem of buffer orchestration in AI data paths by introducing dmaplane, a Linux kernel module that manages buffer allocation, sharing, and lifecycle, resulting in improved performance with measurements showing reduced NUMA cross-node penalties and efficient GPU memory integration compared to cudaMemcpy.
AI transport libraries move bytes efficiently, but they commonly assume that buffers are already correctly allocated, placed, shared, registered, and safe under completion and teardown pressure. This paper presents dmaplane, a Linux kernel module that makes this missing layer explicit as buffer orchestration. dmaplane exposes a stable kernel UAPI via /dev/dmaplane and composes ring-based command channels, DMA buffer lifecycle management, dma-buf export for cross-device sharing, a kernel-space RDMA engine, NUMA-aware allocation and verification, credit-based flow control, low-overhead observability, and GPU memory integration via PCIe BAR pinning. We evaluate orchestration sensitivity with measurements of NUMA cross-node penalties at DRAM scale, completion-safe flow control under sustained RDMA load, and GPU BAR mapping tiers versus cudaMemcpy. We also demonstrate end-to-end disaggregated inference by transferring KV-cache chunks between two machines using RDMA WRITE WITH IMMEDIATE and reconstructing tensor views on the receiver. RDMA measurements use Soft-RoCE; we distinguish measured results from provider-independent properties by construction.