Electrical and Electronic Engineering - Theses

Permanent URI for this collection

Search Results

Now showing 1 - 1 of 1
  • Item
    Thumbnail Image
    Scalable deployment of Video-on-Demand services
    The Video-on-Demand (VoD) service is increasingly becoming popular. Unlike the conventional television services, the VoD service allows viewers to select the videos and the viewing times based on their personal preferences, and also provide the ability to control the video sessions. On one hand, the number of people using VoD service is rapidly increasing day by day. On the other, service providers are deploying many new VoD systems, and also improving the existing systems by enhancing the video libraries and providing videos with better quality. Consequently, the scalability and bandwidth efficiency of VoD systems have become important design considerations today. In addition, the energy efficiency is also now considered as a key concern in the deployment of VoD systems, with the Internet being identified as one of the main potential contributors towards global energy consumption and carbon footprint. In this thesis, we explore the fundamental question of how to realise scalable VoD systems that can sustain the sharp rise of number of VoD users, sizes of VoD libraries, and the quality of video content. In particular, we investigate the strategic exploitation of distributed storage resources to improve the scalability and bandwidth efficiency of VoD systems. To this end, we propose and evaluate architectures, delivery schemes, and techniques, that help VoD systems to efficiently scale and support customers as the number of users, library sizes, and streaming rates increase. In addition, we also analyse the distributed content placement in VoD systems from an energy conservation standpoint. Furthermore, our study also addresses the resulting sub-problems associated with our main objectives, such as estimating popularity of videos and understanding characteristics of VoD users. While unicast is a natural choice for delivering VoD services, unicast-based VoD systems scale very poorly with the system growth. We propose delivery schemes, which improve the scalability and bandwidth efficiency of the VoD systems by strategic exploitation of storage and multicast. The underlying principle behind these schemes is the pre-population of video content, which enables the user requests to be batched and served using multicast. We first use this concept to propose scalable delivery schemes for two-tier VoD systems, where we pre-populate the video content in the end-user's devices. Our simulations based performance evaluations confirm that the proposed schemes significantly improve the scalability and bandwidth efficiency of the VoD systems. We then extend this concept to hierarchical VoD systems and propose a novel scalable VoD delivery scheme, also taking into account the characteristics of VoD users. In this scheme, different portions of the videos are pre-populated in different levels of the network hierarchy. Our trace driven simulations verify the superior performance of our proposed scheme in terms of scalability and bandwidth efficiency. Recently, energy efficiency is also identified as one of the crucial requirements for network based services due to the rising contribution of the Internet towards global energy consumption and carbon footprint. In this thesis, we analyse the distributed content placement in VoD systems from an energy conservation standpoint. To this end, we formulate energy consumption models and numerically analyse those models using data from manufacturers' data sheets. We then use these models to design a replication scheme that improves the energy efficiency of VoD systems by strategic content placement and by switching ON/OFF content storages based on the demand. The popularity of videos is a valuable input for an efficient content placement. We propose a novel method for video popularity estimation in VoD systems. The proposed method uses the arrival times of video requests to derive a parameter called inverted pyramid distance, which captures both long term and short term popularities of videos. Our simulations indicate that the proposed video popularity estimation method performs extremely well despite the popularities of videos being stable or time-varying. The popularity of videos does not capture the personalised interests of an individual user. Therefore, a video popularity based content management strategy might not be the best option for a server that serves a small group of users. We investigate the effectiveness of collaborative filtering (CF) techniques for the content placement at end user's devices. To this end, we propose a bandwidth efficient personalised prefix placement scheme that uses CF techniques, and we analyse it using trace driven simulations. Throughout the thesis, our results provide insight into potential immediate and long term strategies that can significantly improve the scalability and/or energy efficiency of the VoD systems. We also point out possible future research, which will extend the work covered in this thesis while keeping up with the current technology trends.