OpenExpress: Fully Hardware Automated Open Research Framework for Future Fast...
summary
OpenExpress is an open RTL framework that implements NVMe queue, PRP parsing and MSI handling entirely in logic and runs on an UltraScale FPGA with 4×64 GB DDR4, achieving fully hardware automated NVMe processing with no software intervention. The prototype sustains about ~7 GB/s bandwidth (2.8–7 GB/s observed), handles rich outstanding queues and PRP traversal in hardware, and shows lower latencies than a commercial Optane SSD for read‑heavy traces while remaining an FPGA research vehicle.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:33Emerging Non-Volatile Memory for SSDs
- 2:26NVMe Internals and Interfaces
- 3:33NVMe Storage Stack Redesign
- 4:38What Does SSD's CPU Do?
- 7:00Multi-core IP for High-Performance SSD
- 7:27Component Latency Decomposition
- 9:27Overview
- 10:45Queue Dispatching
- 11:17Data Transferring
- 11:38Completion Handling
- 12:02NVMe Context Management
- 12:22Operation Flow of OpenExpress
- 14:32Major IP Cores
- 15:10Synthesis and Implementation
- 15:48Frequency Tuning (50MHz 250 MHz)
- 17:27Perf. Improvement of Frequency Tuning
- 19:39Performance w/ Microbenchmarks
- 21:30Real Workload Evaluation