LPNS: Scalable and Latency-Predictable Local Storage Virtualization for...
summary
LPNS is an OS-level NVMe virtualization that provides latency‑predictable QoS for hybrid local cloud storage by implementing self‑feedback control, flexible I/O queue and command scheduling, scalable polling, and a deterministic network‑calculus model that yields latency upper bounds. It runs as a Linux kernel module (no SmartNICs or FPGAs), is evaluated on Intel Optane and Samsung NVMe SSDs with microbenchmarks and YCSB/RocksDB, and reports up to 18.72× latency optimization, up to 1.45× additional throughput, and a practical 15 ms latency cap with low overhead.
AI-written summary · about