I/O Completion Lab

PAS Algorithm → Simulation/Emulation → Kernel Lab

2026-1 Systems Technology

Lab Overview

This lab provides a step-by-step exploration of NVMe SSD I/O completion techniques. First, understand the PAS (Per-I/O Adaptive Sleep) algorithm, then observe its behavior through simulators/emulators, and finally compare real kernel-level performance in a QEMU virtual machine.

Paper: DPAS: A Prompt, Accurate and Safe I/O Completion Method for SSDs (USENIX FAST '26)

Part 1

PAS Algorithm Introduction

Compare four I/O completion techniques (INT, CP, LHP, PAS) and learn PAS's binary feedback-based adaptive sleep algorithm.

Concepts Paper

Part 2

PAS Simulator / Emulator

Understand the difference between simulator (tracking) and emulator (feedback loop), observe PAS convergence and oversleep effects.

Simulator Emulator Interactive

Part 3

LHP Emulator

Understand Linux Hybrid Polling's epoch-based sleep prediction and the latency shelving problem. Compare with PAS.

lhp_emulator.html Interactive

Part 4

DPAS Introduction & Kernel Lab

Understand PAS limitations and DPAS's runtime mode switching (PAS/CP/INT), then benchmark on a QEMU VM.

DPAS State Machine QEMU Kernel 5.18 bench.sh

References

DPAS Paper (USENIX FAST '26)
DPAS Artifact Repository (GitHub)
dpas.patch Improvement Process — QEMU fixes and AI tool case study

QEMU VM Setup

Choose the guide for your host OS.