PhD Oral Exam - Sepideh Malek Taji, Information and Systems Engineering
Machine Learning for Next-generation Content Delivery Networks: Deployment, Content Placement, and Performance Management
This event is free
School of Graduate Studies
When studying for a doctoral degree (PhD), candidates submit a thesis that provides a critical review of the current state of knowledge of the thesis subject as well as the student’s own contributions to the subject. The distinguishing criterion of doctoral graduate research is a significant and original contribution to knowledge.
Once accepted, the candidate presents the thesis orally. This oral exam is open to the public.
With the explosive demands for data and the growth in mobile users, content delivery networks (CDNs) are facing ever-increasing challenges to meet end-users quality-of-experience requirements, ensure scalability and remain cost-effective. These challenges encourage CDN providers to seek a solution by considering the new technologies available in today’s computer network domain. Network Function Virtualization (NFV) is a relatively new network service deployment technology used in computer networks. It can reduce capital and operational costs while yielding flexibility and scalability for network operators. Thanks to the NFV, the network functions that previously could be offered only by specific hardware appliances can now run as Virtualized Network Functions (VNF) on commodity servers or switches. Moreover, a network service can be flexibly deployed by a chain of VNFs, a structure known as the VNF Forwarding Graph or VNF-FG. Considering these advantages, the next-generation CDN will be deployed using NFV infrastructure. However, using NFV for service deployment is challenging as resource allocation in a shared infrastructure is not easy. Moreover, the integration of other paradigms (e.g., edge computing and vehicular network) into CDN will compound the complexity of content placement and performance management for the next-generation CDNs. In this regard, due to their impacts on final service and end-user perceived quality, the challenges in service deployment, content placement, and performance management should be addressed carefully. In this thesis, advanced machine learning methods are utilized to provide algorithmic solutions for the abovementioned challenges of the next generation CDNs.
Regarding the challenges in the deployment of the next-generation CDNs, we propose two deep reinforcement learning-based methods addressing the joint problems of VNF-FG’s composition and embedding, as well as function scaling and topology adaptation. As for content placement challenges, a deep reinforcement learning-based approach for content migration in an edge-based CDN with vehicular nodes is proposed. The proposed approach takes advantage of the available caching resources in the proximity of the full local caches and efficiently migrates contents at the edge of the network. Moreover, for managing the performance quality of an operating CDN, an unsupervised machine learning anomaly detection method is provided. The proposed method uses clustering to enable easier performance analysis for next-generation CDNs. Each proposed method in this thesis is evaluated by comparison to the state-of-the-art approaches. Moreover, when applicable, the optimality gaps of the proposed methods are investigated as well.