Technical Case Study — Contract Project 2022
DDoS Prediction and Mitigation System
A defensive system based on behavioral analysis and AI traffic classification, delivered under contract for "Nuqtat Ibda" company (July 2022).
Executive Summary
Distributed denial-of-service attacks flood infrastructure with volumes of traffic that are hard to distinguish from legitimate traffic spikes. Traditional threshold-based solutions either react too slowly or block real users. The system solved this by learning the normal traffic baseline, detecting anomalies early, and applying a graduated response.
Project card: Private contract — "Nuqtat Ibda" company, July 2022. Role: design, development, delivery, operational documentation. Status: delivered to client — source code owned by the client.
The Problem
- Static thresholds cannot reliably tell an attack from a legitimate spike.
- Distributed attacks arrive from thousands of sources, each with low individual rates.
- Blanket blocking punishes legitimate users and achieves the attacker's goal.
Methodology: Five Phases
| Phase | Function | Mechanism |
|---|---|---|
| 1 — Baseline | Learn normal traffic patterns | Temporal statistical aggregation |
| 2 — Behavioral detection | Early anomaly detection | Composite behavioral analysis model |
| 3 — Classification | Separate legitimate from attack traffic | Behavioral fingerprints (timing, frequency, path diversity) |
| 4 — Graduated response | Deter without punishing users | Escalating challenges, rate limiting, confirmed blocking |
| 5 — Reporting | Operational visibility | Logs, alerts, post-incident summaries |
Technical Stack
- Python: Processing core and behavioral model learning.
- Redis/Cache: Real-time rate tracking.
- AI Classifier: Traffic classification model.
- Logging & Alerting: Logs, alerts, and operational dashboard.
Testing Methodology
- Simulated L3/L4 and L7 attacks at varying rates.
- Measured false-block rate during traffic peaks.
- Tested response speed to sudden surges.
Lessons Learned
- Composite behavioral detection is more accurate than any single threshold.
- Graduated response protects the user experience.
- Periodic baseline refresh is essential to avoid false alarms.