The Doctoral School in Sciences and Engineering is happy to invite you to Raphaël Charles-Victor OLLANDO’s defence entitled
AI-Enabled Test Data and Schedule Generation Methods for Complex Network Systems
Supervisor: Dr. Seung Yeob SHIN
In our interconnected world, complex network systems are foundational to critical applications like telecommunications, Software-Defined Networking (SDN), and satellite systems. These network systems, with their intricate interdependencies and dynamic interactions, require rigorous testing to ensure reliability, performance, and compliance with industry standards. However, testing complex network systems entails significant challenges arising from their large scale, heterogeneity, and dynamic nature.
Effective testing strategies for complex network systems must generate realistic test data, simulate traffic patterns, and introduce controlled faults to evaluate their fault tolerance and recovery mechanisms. Additionally, efficient test scheduling and resource management are crucial for comprehensive coverage and optimal resource utilization. Hence, in this dissertation, we introduce AI-enabled methods for generating test data and scheduling tests in complex network systems, specifically focusing on SDNs and satellite systems.
Abstract
In Chapter 3, we present FuzzSDN, a machine learning-guided fuzzing method for testing SDN controllers. FuzzSDN efficiently explores the test input space, generating test data that leads to system failures and learning failure-inducing models. FuzzSDN has shown a significant increase in fault detection rates and improved the diagnosis of system failures by providing interpretable models of failure-inducing conditions.
In Chapter 4, we present SeqFuzzSDN, a learning-guided fuzzing method for testing stateful SDN controllers. SeqFuzzSDN leverages the architecture and protocols of SDNs to test controllers in realistic operational settings. It employs Extended Finite State Machines (EFSMs) to guide its fuzzing step, thus resulting in effective and diverse tests that discover failures. Our results demonstrate that SeqFuzzSDN generates more diverse message sequences leading to failures within the same time budget and produces more accurate failure-inducing models, significantly outperforming other methods in terms of sensitivity.
Finally, we present in Chapter 5 a multi-objective approach for scheduling acceptance tests for mission-critical satellite systems. Using the Non-dominated Sorting Genetic Algorithm III (NSGA-III), this approach finds near-optimal feasible schedules that balance operational cost, fragmentation, and resource efficiency. This method has improved the overall efficiency of test campaigns, reducing costs and ensuring thorough testing of satellite systems, while allowing engineers to perform trade-off analyses.