Most research in the area of designing and implementing continuous media servers, which support delay-sensitive data like audio and video, has assumed either clients with very small memory sizes for buffering and no secondary storage (thin clients) or a homogeneous client environment where all clients have exactly the same performance characteristics. Both assumptions are radically changing due to the availability of inexpensive storage, as well as the great diversity of clients that exist now and will exist in the future. In this work we look more closely at the implications that the existence of clients with diverse performance characteristics may have in the server design. Our approach is experimental. We use conventional hardware for servers and clients and examine bottlenecks and optimization options systematically, in order to reduce jitter and increase the maximum number of clients that the system can support. We show that the diversity of client performance characteristics can be taken into account, so that all clients are well supported for delay-sensitive retrieval in a heterogeneous environment. We also show that their characteristics can be exploited to maximnize server throughput under server memory constrains.
展开▼
