DPAS: A Prompt, Accurate and Safe I/O Completion Method for SSDs
summary
DPAS introduces PAS, which estimates sleep by asking whether the last IO woke early or late using only the two most recent completions, and a per-core state machine that dynamically switches among polling, PAS, and interrupts. Implementation is a small block‑layer patch with no dedicated polling thread. PAS reduced CPU usage by 21 percentage points versus Linux hybrid polling for 4 KB random reads, and DPAS improves YCSB performance by 9% on 3D XPoint and 5% on TLC NAND, winning on 11 of 12 SSDs tested.
AI-written summary · about
chapters (11)
- 0:00DPAS introduction and motivation
- 0:40IO completion importance for SSDs
- 2:48Polling limitations and hybrid issues
- 5:40Pass: latency tracking mechanism
- 8:15Deep Pass: dynamic mode switching
- 9:57Implementation and setup
- 10:38Microbenchmark results and CPU gains
- 11:22Database performance under contention
- 12:03Generality across SSDs
- 12:25DPAS contributions and model
- 13:07Summary and Q&A
related talks
13mEnabling DDR segments on demand during memory pressure for DDR power reduction
29mDC-Store: Eliminating Noisy Neighbor Containers using Deterministic I/O Performance
28mXDP for TSN
17mDSA-2LM: A CPU-Free Tiered Memory Architecture with Intel DSA
22mAccelerate the Image Distribution for DADI Using Intel DSA
17mPara-ksm: Parallelized Memory Deduplication with Data Streaming Accelerator
52mDAMON, DAMOS, and DAMO: Kernel Subsystems and User-Space Tools for Data Access...
31mDesign and implementation of Autocaching for CXLSSD