How pushpitarahmank-max/os-term-paper works
This repository is an experimental study of learning-augmented page replacement. Its principal workflow generates a shifted memory-reference trace, trains eviction predictors from an offline Belady oracle, compares replacement policies in simulation, and reports performance and explanation-calibration results. The graph includes the documented policy families and experiment outputs. Source excerpts establish the main orchestration and explanation calls; some subsystem details are documented in the README rather than shown in sampled implementation.
Main entry points: simulator.py Comparative simulation · run_experiments.py Experiment runner · models.py Logistic regression · algorithms.py Learned policies
Components 12
Workload generation
- Shifted trace generator
src/workload.py
Prediction and policies
- Logistic regression
src/models.py - Decision tree
src/models.py - FIFO, LRU, Belady
src/algorithms.py - Learned policies
src/algorithms.py - Adaptive hybrid
src/algorithms.py
Simulation
- Oracle training data
src/simulator.py - Comparative simulation
src/simulator.py
Evaluation and explanations
- Experiment runner
experiments/run_experiments.py - Eviction explanations
src/explanations.py - Calibration metrics
src/explanations.py - Metrics and plots
experiments/run_experiments.py
Connections 13
- Experiment runner → Shifted trace generator: generates traces
experiments/run_experiments.py - Experiment runner → Comparative simulation: runs simulations
experiments/run_experiments.py - Comparative simulation → Oracle training data: extracts labels
src/simulator.py - Comparative simulation → Logistic regression: trains and uses
src/simulator.py - Comparative simulation → Decision tree: trains and uses
src/simulator.py - Comparative simulation → FIFO, LRU, Belady: compares policies
src/simulator.py - Comparative simulation → Learned policies: compares policies
src/simulator.py - Comparative simulation → Adaptive hybrid: compares policy
src/simulator.py - Learned policies → Logistic regression: scores candidates
src/algorithms.py - Adaptive hybrid → Logistic regression: scores candidates
src/algorithms.py - Comparative simulation → Eviction explanations: generates explanations
src/simulator.py - Comparative simulation → Calibration metrics: computes calibration
src/simulator.py - Experiment runner → Metrics and plots: writes reports
experiments/run_experiments.py