Scheduling strategies for Interactive Video-On-Demand service

This reseach project is a theoretical and experimental investigation of scheduling strategies that can improve the performance of Interactive Video-On-Demand (IVOD) servers. Examples of IVOD can be found in digital libraries, hypermedia, distance learning, entertainment and telemarketing applications. In contrast to the delivery of movies, interactive video programmes are seldom linear. IVOD programmes usually consist of short video branches separated by user interactions. IVOD users will not be satisfied with long latency for start-up and restart of video branches. Further, interactive media may be accompanied by other forms of data such as voice, audio, image, text or graphics. For that reason, an IVOD server also needs to provide adequate residual bandwidth to support sporadic data requests arising from user interactions. Therefore, an IVOD scheduler needs to address performance issues such as startup latencies, queuing time, admission probabilities, and data throughput. As the length of video branches in IVOD are typically short, departures and startup of requests are frequent. The adverse effects of these frequent transients on throughput and robustness must also be treated explicitly. The user-driven interactivity in IVOD service poses scheduling problems that have not been fully addressed by the admission control and static disk-scheduling techniques previously proposed for VOD servers. In this research, we identify IVOD performance issues and present solutions to address some of these performance issues. Our philosophy is to dynamically schedule request admissions and disk accesses in an IVOD server, by exploiting the use of run-time information. At the disk scheduling level, we introduce two disk scheduling strategies. To minimize start-up latency and achieve good tolerance against variable-bit-rate, we introduce the GS_EDF scheduler. To reduce the long start-up latency inherent to seek-reducing disk schedulers, we introduce the GSCS scheduler which can significantly reduce start-up latency while preserving a high stream throughput. At the admission level, we introduce admission prioritizing strategies which can improve queuing time and total throughput of a busy IVOD server without causing undesirable throughput degradation and bandwidth fragmentation. To improve sporadic service throughput without losing streaming throughput, we introduce an optimum readsize control technique to maximize the disk efficiency dynamically. The readsize control technique is also novel in removing disk rotational latencies. All the techniques introduced in this thesis are verified by comprehensive performance evaluations. The results confirm that these proposed strategies can improve IVOD service in various dimensions of performance