Yongsoo Joo

DPAS: I/O Completion Optimization for SSDs USENIX FAST 2026

PAS Sleep Duration Tracker

Simulates how the PAS (Per-I/O Adaptive Sleep) algorithm tracks I/O latency and dynamically adjusts sleep duration using binary feedback (UNDER/OVER). Visualizes the convergence behavior with various I/O trace patterns.

Adjustment rate tuning, rule frequency analysis, and convergence tracking

PAS I/O Completion Emulator

Emulates how sleep misprediction affects application-perceived I/O latency. Shows the interplay between genuine device latency, requested sleep, actual wake time (including context-switch overhead), and the resulting perceived latency.

Sleep misprediction and app-perceived latency visualization

LHP I/O Completion Emulator

Demonstrates the "latency shelving" problem in the LHP (Latency-based Hybrid Polling) method, where oversleep causes inflated latency measurements that feed back into higher sleep predictions. Useful for comparing against PAS/DPAS behavior.

Latency shelving visualization and comparison baseline for DPAS

Source code: github.com/DongDongJu/DPAS_FAST26