Heuristic batching policies for video-on-demand services

Abstract

A video-on-demand (VOD) service imposes extremely severe resource requirement in terms of bandwidth and storage. Batching policies that use a single channel to serve multiple active clients for the same video program can reduce system resource requirement effectively, especially during high load periods. In this paper, we study heuristic batching policies for VOD systems without a priori knowledge of traffic pattern. In waiting VOD systems without reneging due to long latency for playback requests, the average latency time is an important performance index about the service quality. For such VOD systems without reneging, we propose a novel batching scheme, called Statistical MFQL ( Maximum Factored Queue Length), to reduce the average latency time for playback requests. Our scheme performs better than the Instantaneous MFQL batching scheme proposed earlier. In loss VOD systems with reneging due to long latency, the reneging ratio becomes an important performance index about the system revenue. For such VOD systems with reneging, we propose another novel batching scheme, called Max-Cost-First, to reduce the reneging ratio for playback requests. We also study the impact of channel allocation schemes upon these novel batching schemes. According to our simulation results for various workloads, the simple Regular-Interval channel allocation scheme performs better than the proposed Rate-Based scheme for cooperating with either the Statistical MFQL or the Max-Cost-First batching scheme.