Publications & Technical Research

Abstract

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.

Abstract

Cryptographic systems deployed in production often differ substantially from their documented or configured cryptographic posture, creating blind spots that complicate security analysis and post-quantum migration planning. We introduce a dual-layer observability model that distinguishes between configured cryptography (static layer) and observed cryptography (runtime layer), and formalize cryptographic drift as the mismatch between these layers. We define two classes of drift: Configured-Not-Observed (CNO) and Observed-Not-Configured (ONC), and establish basic properties of drift detection under this model.

We present a reference implementation that realizes the model by combining binary-level cryptographic inventorying, runtime tracing of cryptographic operations, and passive network protocol analysis. The implementation enables detection of undocumented libraries, dynamically loaded cryptographic code, and network-visible cryptographic behavior without payload decryption. Experimental results on Linux and embedded systems demonstrate scalable static analysis, production-safe runtime tracing, and high-throughput network observation, and reveal substantial cryptographic drift in real deployments.