Need for a Deeper Cross-Layer Optimization for Dense NAND SSD to Improve Read
Talkat 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage '20) · 2020-07-13..07-14
summary
Proposes combining LSB/CSB/MSB pages in a wordline into a single Melded-pages logical page, implemented in a SimpleSSD simulator with HDFS workloads and scheduler that reorders writes around shadow programming to keep all three subpage reads queued; measures show per-page read latencies (LSB 58μs, CSB 78μs, MSB 107μs), models overhead as x≈40μs and y≈18μs, finds benefits only for reads >2MB and reports up to 45% throughput improvement using an added NVMe directive for read-pattern hints.
AI-written summary · about