Storage Acceleration via Decoupled Software Defined Storage Architecture

Explore a technical conference presentation from SNIA SDC 2024 that delves into accelerating storage performance through a decoupled Software Defined Storage (SDS) architecture. Learn about innovative approaches to overcome performance limitations in SDS frameworks, where traditional millisecond access latencies and limited IOPs throughput can be significantly improved. Discover a novel architecture that bypasses bottleneck SDS stacks by implementing decoupled data read paths, with detailed explanations of how logical-to-physical storage mappings are shared and cached. Follow the implementation of a proof-of-concept in Openstack Ceph using SPDK for direct NVMe-over Fabrics I/O paths, demonstrating remarkable improvements including 10x read latency reduction and 4-10X throughput increases. Master key concepts including I/O performance bottlenecks in SDS architecture, NVMe-oF integration challenges with distributed storage, benefits of user mode storage stacks, I/O path optimization strategies, and methods for measuring storage I/O performance accurately. Presented by Intel's Arun Raghunath, this 49-minute talk provides valuable insights into the future of storage optimization, including CPU usage reduction, hardware assist utilization, and workload-specific optimizations.