Abstract

Multimedia conferencing is the real-time exchange of multimedia content between multiple parties. It is the basis of several applications, such as distance learning, online meetings, and massively multiplayer online games. Cloud-based provisioning of multimedia conferencing has several benefits, like resource efficiency, elasticity, and scalability. However, it remains very challenging. A challenge, for instance, is the lack of holistic architectures which cover both the infrastructure and the platform layers of cloud-based multimedia conferencing applications. Another challenge is the lack of appropriate algorithms for resource allocation in the conferencing cloud to accommodate the fluctuating number of participants, while meeting the required quality of services (QoS). Yet another example is the lack of suitable algorithms for scaling the multimedia conferencing applications in the cloud while meeting both QoS requirements and cost efficiency objective. Unfortunately, the solutions proposed so far do not address these challenges.
This thesis focuses on the architectural and algorithmic challenges of cloud-based multimedia conferencing. It proposes architectural components and interfaces for multimedia conferencing application provisioning, covering both the Platform-as-a-Service (PaaS) and the Infrastructure-as-a-Service (IaaS) layers. The proposed interfaces simplify multimedia conference service provisioning for a wide range of application providers. On the algorithmic side, it proposes resource allocation mechanisms that support scalability in terms of the number of participants while meeting the QoS. These mechanisms allocate the actual resources (e.g., CPU, RAM, and storage) in an optimal manner. Besides these mechanisms, it proposes the scalability approaches for cloud-based multimedia conferencing applications. To ensure cost efficiency, these proposed solutions enable fine-grained scalability of the applications with respect to the number of participants while considering the QoS requirements. All algorithmic problems in this thesis are formulated using the Integer Linear Programming (ILP) and heuristics have been designed and validated to solve them.