CPUXpert — Intelligent CPU Scheduling Simulator

A desktop CPU scheduling simulator featuring a real ML-based algorithm recommender (91.5% accuracy), real system monitoring via psutil, and a 54-test pytest suite verified against OS textbook examples.

---

Features

Scheduling Algorithms — scheduler.py

Six classic scheduling algorithms implemented in the Scheduler class:

Algorithm	Method	Preemptive	Description
FCFS	fcfs()	No	First Come First Served — processes run in arrival order
SJF	sjf()	No	Shortest Job First — picks the shortest burst time among available processes
SRTF	sjf_preemptive()	Yes	Shortest Remaining Time First — re-evaluates at every time unit
Round Robin	round_robin(q)	Yes	Time-sliced with configurable quantum
Priority	priority()	No	Higher value = higher priority
Priority (Pre)	priority_preemptive()	Yes	Preempts when a higher-priority process arrives

ML Algorithm Recommender — ml_recommender.py

A trained Random Forest classifier that predicts the best scheduling algorithm for any workload.

How it works:

2,000 Random Workloads → Run All 6 Algorithms → Label Best → Extract 8 Features → Train Classifier

Training results (printed by python ml_recommender.py, fully reproducible with seed=42):

Model Selected:        Random Forest (200 estimators, max_depth=15)
5-Fold CV Accuracy:    90.88% ± 0.98%
Test Accuracy:         91.50%
Test F1 (weighted):    91.35%
Test F1 (macro):       77.70%

Per-class performance:

Class	Precision	Recall	F1-Score	Support
FCFS	1.00	0.33	0.50	3
SJF	0.94	0.87	0.90	181
SRTF	0.90	0.96	0.93	216

Confusion matrix:

Predicted →   FCFS   SJF   SRTF
Actual FCFS  [  1     2     0  ]
Actual SJF   [  0   157    24  ]
Actual SRTF  [  0     8   208  ]

Feature importances:

Feature	Importance
arrival_spread	31.52%
arrival_var	31.28%
burst_arrival_ratio	15.28%
burst_cv	5.83%
burst_var	5.62%
burst_mean	4.39%
priority_var	3.13%
process_count	2.95%

Label distribution across 2,000 workloads (ground truth — which algorithm actually wins):

Algorithm	Count	Percentage
SRTF	1,083	54.1%
SJF	903	45.1%
FCFS	14	0.7%
Priority / Round Robin	0	0.0%

SRTF and SJF dominate because they minimize waiting time by design. Priority and Round Robin optimize for fairness, not waiting time.

Real System Monitoring — task_manager.py

The Performance Monitor reads real system stats using psutil:

Metric	Source	Unit
CPU	psutil.cpu_percent()	% (0–100)
Memory	psutil.virtual_memory().percent	% (0–100)
Disk I/O	psutil.disk_io_counters() delta	MB/s
Network I/O	psutil.net_io_counters() delta	Mbps

Sampled every 1 second. Plotted as 60-second rolling history. Zero random number generation in the monitoring code.

Process Control — task_manager.py

Action	What it does
Pause	Adds PID to paused_pids; process is skipped in scheduling
Resume	Removes PID from paused_pids; process re-enters scheduling
Terminate	Sets remaining_time = 0 and marks completion_time on the Process object

These modify actual scheduler.Process state — not cosmetic.

---

Test Suite

54 tests, 100% passing — verified with python -m pytest -v:

tests/test_scheduler.py       — 35 tests (algorithm correctness + edge cases + metrics)
tests/test_ml_recommender.py  — 19 tests (workload gen + features + labeler + prediction)

Coverage:

Module	Statements	Covered	Coverage
scheduler.py	239	205	86%
ml_recommender.py	238	131	55%
test_scheduler.py	182	177	97%
test_ml_recommender.py	124	119	96%

Textbook-verified test cases include:

• FCFS: P1(0,24), P2(0,3), P3(0,3) → WT: [0, 24, 27], Avg WT = 17.0
• SJF: P1(0,6), P2(0,8), P3(0,7), P4(0,3) → WT: [3, 16, 9, 0], Avg WT = 7.0
• SRTF: P1(0,8), P2(1,4), P3(2,9), P4(3,5) → WT: [9, 0, 15, 2], Avg WT = 6.5
• Round Robin (q=4): P1(0,24), P2(0,3), P3(0,3) → WT: [6, 4, 7]
• Priority: P1-P5 with priorities 3,1,4,5,2 → order: P4→P3→P1→P5→P2
• Edge cases: single process, all-same-arrival, all-same-burst, priority ties, CPU gaps

CI pipeline (.github/workflows/test.yml): runs on every push/PR across Python 3.10, 3.11, 3.12.

---

Tech Stack

Component	Technology
Language	Python 3.10+
GUI	Tkinter
Visualization	Matplotlib
Numerical	NumPy
Machine Learning	scikit-learn (Random Forest)
System Monitoring	psutil
Model Persistence	joblib
Testing	pytest + pytest-cov
CI/CD	GitHub Actions

---

Quick Start

# Clone
git clone https://github.com/ankan123basu/CPUXpert.git
cd CPUXpert

# Virtual environment
python -m venv .venv
.venv\Scripts\activate        # Windows
# source .venv/bin/activate   # Linux/Mac

# Install
pip install -r requirements.txt

# Train the ML model (first time only — takes ~40s)
python ml_recommender.py

# Run the simulator
python main.py

Running Tests

python -m pytest --cov=. --cov-report=term-missing -v

Running Benchmarks

python benchmark.py

---

Project Structure

CPUXpert/
├── main.py                    # Tkinter GUI application
├── scheduler.py               # 6 scheduling algorithms
├── ml_recommender.py          # ML training pipeline + prediction API
├── task_manager.py            # psutil monitoring + process control
├── benchmark.py               # Algorithm benchmarking
├── requirements.txt           # Dependencies
├── pytest.ini                 # Test configuration
├── .gitignore
├── ml_model/                  # Generated by ml_recommender.py
│   ├── model.pkl              #   Trained Random Forest (200 trees)
│   ├── scaler.pkl             #   StandardScaler
│   └── metadata.json          #   Accuracy, F1, config
├── tests/
│   ├── test_scheduler.py      # 35 algorithm tests
│   └── test_ml_recommender.py # 19 ML module tests
└── .github/workflows/
    └── test.yml               # CI on Python 3.10/3.11/3.12

---

Scope & Limitations

• This is a desktop simulator — not a web app, not deployed to cloud
• The ML model is a Random Forest, not deep learning
• It simulates OS scheduling — it does not schedule real OS processes
• System monitoring shows your machine's real stats but does not control them