Networking and storage support for video-on-demand data delivery
Abstract
Video-on-Demand gives users greater control over what, when and where they watch videos. As the number of subscribers grows, it strains the current network and storage infrastructure since VoD has large bandwidth requirements when compared to other services. Infrastructure upgrade is costly and could lead to wastage of resources if not used fully. To address this problem, we propose solutions that utilize the deployed infrastructure optimally while maintaining the required QoS. This thesis develops mathematical models that seek to capture workload characteristics and provide solutions by optimally solving these models. In the first part, we describe our work on optimizing data retrieval in a disk-array for a VoD back-end server. Data replication is based on popularity metrics like read and write and retrieval uses a linear programming algorithm for serving requests optimally. On a prototype VoD server, our algorithms provide better performance than previously used data-placement policy like RDA with respect to total number of clients supported. Our algorithms perform better than RDA by 14.3% and 12% for workloads that follow a popularity distribution of Zipf with alpha = 0.5 and alpha = 1.0 respectively. In the second part, we propose smart replication policies for pre-seeding data in set-top boxes in a peer-to-peer VoD network. We present a mathematical algorithm which reduces the uplink traffic from the ISP community to the central server by upto 50%. In the last part, we describe our efforts in studying and helping manage 3G congestion using incentives. Through a field-trial of students at Purdue campus, we study smartphone usage behavior of 16 users for several months and then use insights gained from the field study to provide incentives to users to help manage 3G congestion. Our results indicate a high level of compliance with economic incentives and disincentives, further correlated with two psychological measures of each user (agreeableness and neuroticism). Our study provides insights in smartphone usage behavior with respect to network usage like the amount of WiFi and 3G usage per user and the popular applications used along with the popular protocols used to access such applications. The Purdue Mobility Lab setup as part of this work can be reused and allows researchers to conduct any further research easily.
Degree
Ph.D.
Advisors
Pai, Purdue University.
Subject Area
Computer Engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.