GAMR: An Enhanced Non-dominated Sorting Genetic Algorithm II-based Dynamic multi-objective QoS routing in Software-Defined Networks

Routing optimization plays a crucial role in traffic engineering, aiming to efficiently allocate network resources to meet various service requirements. In dynamic network environments, however, network configurations constantly change, therefore single-objective routing faces numerous challenges in managing multiple concurrent demands. Moreover, the complexity of the problem increases as end-to-end Quality of Service (QoS) requirements and conflicts between them accumulate. Several approaches have been proposed to address this issue, but most of them fall into the stability-plasticity dilemma or involve excessive computation or convergence times in practical implementations. \textcolor{blue}{We introduce a dynamic multi-objective QoS routing approach based on NSGA-II (Non-dominated Sorting Genetic Algorithm II), called GAMR, utilizing QoS metrics to construct a multi-objective function.} By proposing new initialization and crossover strategies, our solution can find optimal solutions within a short runtime. Additionally, The GAMR application is deployed on the control plane within Software-Defined Networks (SDNs) and evaluated against benchmark methods under various settings. Compared to multi-objective algorithms, the proposed method demonstrates significant improvements in performance indicators. It shows enhancements ranging from 3.4% to 22.8% on the Hypervolume (HV) metric and from 33% to 86% on the Inverted Generational Distance (IGD) metric. Regarding the objectives, the experimental results demonstrate that our method reduces the forwarding delay and packet loss rate to 41.25ms and 3.9%, respectively, under the most challenging network configuration scenario (with only 2 servers and up to 100 requests) on the Chinanet network (44 nodes).

Requirements

The project requires the following Python libraries:

• ryu
• fastapi[all]
• mininet
• networkx
• numpy
• requests

Project Structure

The project is organized into four main modules:

• dynamicsdn: Contains the NSGA-II implementation and Dijkstra's algorithm for path selection.
• scenario: Stores different scenarios for automated testing.
• sdndb: Provides a database for storing network information and routing results.
• mn_restapi: Implements a REST API hook for interacting with Mininet scripts.

Startup Instructions

1. Set environment variable: Before starting the application, add the parent directory of the project to your Python path. In your terminal, run:

export PYTHONPATH={parent directory of project}

2. Start application: Run the following command to start both Ryu and the application:

./startup.py

Further Information

• For detailed documentation, please refer to the project's internal documentation.
• Feel free to open issues on GitHub for any questions or feedback.