AI Routing Lab

Predictive Route Selection using Machine Learning for Latency/Jitter Optimization

Available Documentation:

• English: This document (README.md)
• Russian: README_ru.md - Русская документация проекта

Project Overview

AI Routing Lab is a research project focused on developing machine learning models for predictive route selection in CloudBridge network infrastructure. The project aims to achieve >92% accuracy in predicting latency and jitter for optimal route selection.

Key Objectives:

• Predictive route selection based on latency/jitter prediction
• Integration with quic-test for model validation on real QUIC traffic
• Production integration with CloudBridge Relay for real-time routing optimization

About CloudBridge Research

Autonomous Non-Commercial Organization

This project is part of CloudBridge Research Center — an independent research center specializing in network technologies, distributed systems, and cybersecurity.

Organization:

• Conducts fundamental and applied research in network protocols (QUIC, MASQUE, BGP, etc.)
• Develops and distributes open-source software
• Provides educational programs and training
• Collaborates with leading universities and research institutions
• Prepares highly qualified specialists for industry

Contact & Resources:

• Website: https://cloudbridge-research.ru/
• GitHub: CloudBridge Research
• Email: info@cloudbridge-research.ru

Research Goals

Primary Goal

Develop ML models that can predict route latency and jitter with >92% accuracy to enable proactive route selection in CloudBridge network.

Research Areas

1. Latency Prediction

   • Time-series forecasting of route latency
   • Multi-path latency comparison
   • Historical pattern analysis

2. Jitter Prediction

   • Jitter variability modeling
   • Network condition impact analysis
   • Route stability assessment

3. Route Selection Optimization

   • Ensemble models for route ranking
   • Real-time prediction inference
   • Integration with CloudBridge Relay

Architecture

┌─────────────────────────────────────────────────────────┐
│              AI Routing Lab (Python)                    │
│                                                         │
│  ┌──────────────────────────────────────────────────┐   │
│  │  Data Collection                                 │   │
│  │  • Prometheus metrics from quic-test             │   │
│  │  • JSON export from quic-test                    │   │
│  │  • Historical data storage                       │   │
│  └──────────────────────────────────────────────────┘   │
│                          │                              │
│                          ▼                              │
│  ┌──────────────────────────────────────────────────┐   │
│  │  ML Pipeline                                     │   │
│  │  • LatencyPredictor (Random Forest)              │   │
│  │  • JitterPredictor (Random Forest)               │   │
│  │  • RoutePredictionEnsemble                       │   │
│  │  • Feature engineering                           │   │
│  │  • Model evaluation                              │   │
│  └──────────────────────────────────────────────────┘   │
│                          │                              │
│                          ▼                              │
│  ┌──────────────────────────────────────────────────┐   │
│  │  Inference Engine                                │   │
│  │  • Real-time predictions                         │   │
│  │  • Route optimization                            │   │
│  │  • API for CloudBridge Relay                     │   │
│  └──────────────────────────────────────────────────┘   │
│                          │                              │
│                          │ Validation                   │
│                          ▼                              │
│  ┌──────────────────────────────────────────────────┐   │
│  │  quic-test (Go)                                  │   │
│  │  • Real QUIC traffic generation                  │   │
│  │  • Metrics collection                            │   │
│  │  • ML prediction validation                      │   │
│  └──────────────────────────────────────────────────┘   │
└─────────────────────────────────────────────────────────┘

Project Structure

ai-routing-lab/
├── README.md                    # This file
├── LICENSE                      # MIT License
├── requirements.txt             # Python dependencies
├── setup.py                     # Package setup
│
├── data/                        # Data collection and processing
│   ├── collectors/
│   │   └── quic_test_collector.py    # quic-test integration
│   └── pipelines/
│
├── models/                      # ML model definitions
│   ├── core/                    # Core ML infrastructure
│   │   ├── model_registry.py    # Model versioning
│   │   ├── data_preprocessor.py  # Data preprocessing
│   │   └── feature_extractor.py # Feature engineering
│   ├── prediction/              # Prediction models
│   │   ├── latency_predictor.py # Latency prediction
│   │   ├── jitter_predictor.py  # Jitter prediction
│   │   └── route_prediction_ensemble.py # Route selection
│   ├── routing/                 # Route optimization models
│   ├── anomaly/                 # Anomaly detection (optional)
│   └── monitoring/              # Model monitoring (optional)
│
├── training/                    # Training scripts
│
├── inference/                   # Inference engine
│
├── evaluation/                  # Model evaluation
│
├── experiments/                 # Laboratory experiments
│   ├── lab_experiment.py       # Experiment framework
│   ├── example_experiment.py   # Example experiment
│   └── latency_jitter_experiment.py # Complete workflow
│
└── docs/                        # Documentation
    ├── ARCHITECTURE.md          # Architecture documentation
    └── INTEGRATION_GUIDE.md     # Integration guide

Quick Start

Prerequisites

• Python 3.11+
• quic-test running and exporting metrics
• Prometheus (optional, for metrics collection)

Installation

Basic Usage

Models

LatencyPredictor

Random Forest model for predicting route latency.

Features:

• Historical latency patterns
• Route characteristics (PoP locations, BGP paths)
• Network conditions (congestion, packet loss)
• Time-based features

Target: >92% accuracy (R² score)

JitterPredictor

Random Forest model for predicting route jitter variability.

Features:

• Historical jitter patterns
• Route stability metrics
• Network variability indicators

Target: >92% accuracy (R² score)

RoutePredictionEnsemble

Combines latency and jitter predictions for optimal route selection.

Scoring:

• Latency weight: 70%
• Jitter weight: 30%
• Selects route with best combined score

Target: >95% optimal route selection

Integration with quic-test

AI Routing Lab integrates with quic-test for:

1. Data Collection:

   • Prometheus metrics export from quic-test
   • JSON export for historical data
   • Real-time metrics streaming

2. Model Validation:

   • Validate ML predictions against real QUIC traffic
   • Compare predicted vs actual latency/jitter
   • Calculate prediction accuracy metrics

3. Production Testing:

   • Test route selection in controlled environment
   • A/B testing framework
   • Performance benchmarking

Setup Integration

1. Start quic-test with Prometheus export:

2. Collect metrics:

Laboratory Experiments

The project includes a comprehensive laboratory experiment framework:

See

Documentation

Architecture

• Architecture Documentation
• Integration Guide

Guides

• Usage Examples
• Contributing Guide

Development

• Adaptation Status
• Adaptation Notes
• Upgrade Summary

Experiments

• Laboratory Experiments

Reports

• Laboratory Reports - Test reports organized by date and version

Technology Stack

• Language: Python 3.11+
• ML Framework: scikit-learn (Random Forest), TensorFlow/PyTorch (optional)
• Experiment Tracking: MLflow
• Data Processing: pandas, numpy
• Metrics Collection: prometheus-client
• API: FastAPI / gRPC

License

This project is licensed under the MIT License - see the LICENSE file for details.

Related Projects

• quic-test - QUIC protocol testing tool
• CloudBridge Relay - Production relay server
• CloudBridge Research - Research center

Contact

• GitHub: @twogc
• Email: info@cloudbridge-research.ru
• Website: cloudbridge-research.ru

Acknowledgments

This research project is part of CloudBridge Research Center and integrates with the quic-test testing framework.

Models and infrastructure adapted from CloudBridge AI Service ecosystem.

Status: In Active Development
Last Updated: November 2025