Technical Case Study — Contract Project 2022

DDoS Mitigation System

A defensive system based on behavioral analysis and AI traffic classification.

Year2022
RoleContract project — design, development, delivery
StatusDelivered to client — source owned by client

Project Summary

A contracted system designed to distinguish legitimate traffic spikes from distributed attacks by learning a behavioral baseline and applying a graduated response that protects user experience.

DDoS ProtectionBehavioral AnalysisAI

The Problem

Traditional DDoS defence works with a fixed threshold: if traffic crosses a number, start blocking. The problem is that a smart attack produces traffic that looks legitimate and crosses the threshold quietly — and once you block, you lose real visitors along with the attacker.

What is needed is a system that reads the behaviour of traffic, not only its volume.

Key Features

Behavioral baseline

Learns normal traffic patterns over time.

Early detection

Detects anomalies before they saturate capacity.

Behavioral fingerprints

Separates users from bots via timing, frequency, and path diversity.

Graduated response

Escalating challenges, rate limits, then confirmed blocks.

Operational reports

Alerts and post-incident summaries.

Technical Challenges & Design Decisions

  • A moving baseline: normal traffic differs by hour and by day. The system's first stage is therefore not detection but learning the baseline, refreshed on a schedule.
  • Graduated response: full blocking is the last option, not the first. The sequence is escalating verification, then rate limiting, then confirmed blocking — protecting real users even mid-attack.
  • Decision speed: a two-second-late decision means an outage, so analysis runs in the traffic path rather than after it.

Technical Architecture

Five-stage methodology — baseline is learned first, deviation is then detected, response escalates gradually
منهجية خمس مراحل — تُتعلَّم خط الأساس أولًا، ثم يُكشف الانحراف، وتتدرّج الاستجابة
PythonProcessing core and model learning.
Redis/CacheReal-time rate tracking.
AI ClassifierTraffic classification model.
Logging & AlertingLogs, alerts, and operational dashboard.

Testing Methodology

  • Simulated layer 3/4 and layer 7 attacks at varying rates and measured the false-block rate.
  • Measured response time to the first abnormal spike.
  • Tested a legitimate traffic spike (marketing campaign) to confirm it is not classified as an attack.

Outcome & Impact

  • Attacks distinguished from genuine traffic spikes (a campaign, a breaking story) instead of treating both the same.
  • Legitimate users stay served during mitigation.
  • Post-incident reporting: what happened, when, and from where.

Contract project (July 2022) delivered to Nuqtat Ibdaa — read the full case study.