Supporting video on demand

We examine I/O scheduling strategies for maximizing the number of concurrent streams that can be supported by a VOD system implemented using current midrange device technologies. The overall VOD system is modeled as a collection of resources, characterized by their bandwidths and capacities. We determine limits on the maximum number of concurrent streams possible for various I/O subsystem configurations. We then propose several practical I/O scheduling algorithms and determine the amount of memory required at both the disk and server to support the limiting number of concurrent streams. An admission control policy is proposed for limited memory environments. Finally, we discuss client request (workload) distributions and data placement strategies and their effects on the number of concurrent streams and memory buffer requirements. If the client video requests are not badly skewed (as quantified below) and if sufficient memory is available on both the disk on-board cache and server, our practical algorithms can support the maximum number of concurrent streams allowed by the I/O subsystem configuration