DSA-2LM - A CPU-Free Tiered Memory Architecture with Intel DSA

Learn about a novel tiered memory architecture that leverages Intel's Data Streaming Accelerator (DSA) to optimize heterogeneous memory management in this 17-minute conference presentation from USENIX ATC '25. Discover how DSA-2LM addresses the critical trade-offs between optimal data placement and costly data movement in systems using Persistent Memory or CXL Memory by implementing CPU-free hardware acceleration that achieves up to 4× faster data movement compared to single CPU cores. Explore the technical challenges of fine memory movement granularity in Linux kernel that limit performance improvements and understand how the proposed framework integrates fast memory migration workflows with adaptable concurrent data paths and well-tuned DSA configurations. Examine experimental results demonstrating 20%, 30%, and 16% performance improvements over existing tiered memory solutions MEMTIS, TPP, and NOMAD respectively when tested with real-world applications, presented by researchers from Tsinghua University, University of Electronic Science and Technology of China, Alibaba Group, and The University of Texas at Arlington.