Support for service scalability in video-on-demand end-systems.
Emmanuel L. Abram-Profeta, Kang Geun Shin · Deep Blue (University of Michigan) · 1998
Large-scale deployment of video-on-demand (VoD) systems depends on the quality-of-service (QoS) in customers' interactions. These interactions typically involve the VoD end-systems, that is, the local VoD server and customers' premise equipment (CPE), which communicate via an access network. At this level, factors such as storage reconfigurations and time variations in the request arrival pattern usually cause customers' QoS degradation. This dissertation determines how to achieve service scalability, which is defined as the ability to accommodate possibly significant workload fluctuations and overloading situations without affecting customers' QoS in interactions. First, VoD server storage should be organized to (1) make a large collection of movie titles and concurrent channels available at the lowest possible cost, and (2) offer the lowest admission latency, or minimum waiting time for service. We show that the coarse-grained striping scheme in disk arrays is a potential candidate for its cost-effective storage usage and low admission latency. In addition, we identify the feasibility of clustered disk-array-based storage organizations to reduce service disruptions during content updates. For a given storage organization, scalability can be further improved by batching customers' requests and using multicast communication. For instance, in Near-VoD (NVoD), videos are sourced at equally-spaced intervals, thus allowing limited VCR functionality and guaranteeing a specified maximum admission latency. In order to fully exploit these features, an analytical study of NVoD is presented, which compares customers' and service provider's conflicting objectives. If, unlike in NVoD, batching is done without any control in the channel allocation process, extended periods of demand surge may lead to congestion cycles, causing a scalability problem. We thus introduce a methodology to measure scalability and identify scalable alternatives to NVoD. These multicast VoD systems are applicable in a wide range of clustered storage organizations, and may offer a better tradeoff between customers' QoS and server's throughput. However, in these multicast VoD systems, continuity in VCR actions can only be provided intermittently. To alleviate this limitation, we propose and evaluate mechanisms which greatly improve customers' VCR functionality with minimal degradation of service scalability.