| Recent advances in communication and storage technologies have spurred a strong interest in Multimedia-on-Demand (MOD) services. These services eliminate the short-comings of their broadcast-based counterparts by providing customers with convenience, choice, and control.; The purpose of this thesis is to address the two principal challenges facing multi-media servers: supporting large numbers of concurrent customers and scaling easily and cost-effectively in order to cope with the increasing demands. The number of customers that can be serviced concurrently is highly constrained by the stringent requirements of the real-time playback and the high transfer rates. These requirements lead to rapid consumptions of server resources, especially the storage subsystem bandwidth and the network bandwidth. The scalability challenge arises primarily because current multimedia servers are based on the conventional fileserver architecture, where all the data flowing from the disks to the clients pass through the server. The required store-and-forward copying imposes unnecessary burdens on these servers, severely limits their scalability, and increases their costs.; First, this thesis proposes an Adaptive Block Rearrangement policy that enhances the performance of multimedia storage subsystems by dynamically rearranging the blocks on each disk according to their access frequencies. The effectiveness of the adaptive policy and the impacts of different data placement layouts on performance are demonstrated through extensive simulation.; Second, this thesis proposes a new class of scheduling policies, called Next Schedule Time First (NSTF), which provides hard time of service guarantees and accurate schedule times. Providing such guarantees not only enhances customer-perceived quality of service (QoS) but also increases server throughput by motivating customers to wait. The thesis presents alternative implementations of NSTF and demonstrates through extensive simulation that NSTF delivers outstanding performance benefits.; Third, this thesis develops a validated analytical model for caching in multimedia servers. The model is then used to examine the impacts of various system and workload parameters and to experiment with new design alternatives.; Fourth, this thesis proposes using the emerging Network-Attached Disk (NAD) architecture for designing large-scale, cost-effective, and scalable multimedia servers. The NAD architecture removes all store-and-forward copying, which severely limits the scalability and increases the costs of conventional multimedia servers.; Finally, this thesis develops an Integrated Resource Sharing policy, which improves the performance of NAD-based multimedia servers through caching and intelligent scheduling. This policy combines two schemes that are also proposed here: Distributed Interval Caching (DIC) and Multi-Objective Scheduling (MOS). The DIC scheme exploits the internal caches of NADs in caching intervals between successive streams. The MOS scheme schedules the waiting requests intelligently based on four performance criteria. The integrated policy increases the number of serviced customers while reducing their waiting times for service, uses server and network resources efficiently, does not expect much resource from customers, achieves greater performance benefits for larger servers, and provides high configurability. The effectiveness of the integrated policy and the interactions among various system and workload parameters are studied through extensive simulation. The performance limit of DIC is also evaluated using analytical modeling. |
