|Abstract:||Multimedia services such video-on-demand service to large number of clients require stable network bandwidth provision and short network delay. In order to achieve these goals, the object replication in overlay P2P networks in a good candidate solution for both service providers and customers. However, the characteristics of peers are different from traditional content distribution infrastructure, i.e., dynamic membership changes, asymmetric network bandwidth utilization, node heterogeneity, therefore dynamic change of service availability, etc., so that they make the multimedia content distribution more challenging. In this thesis, I present the collaborative object placement and replication schemes in collaborative computing environments with different scale of group size.
In cooperative streaming and caching, we present a QoS-aware middleware for collaborative multimedia caching and caching service in heterogeneous computing environment, which not only reduces the initial delay of playing time, but also minimizes the delay jitter during playing time and net- work bandwidth utilization. With the publish/subscribe mechanism of cache service, we shows collaborative techniques for multimedia streaming, streaming scheduling and pre-fetching embedded in the QoS middleware architecture for dynamic group management.
In distributed caching, we propose a novel distributed caching algorithm for multimedia streaming, supporting a large number of peers over P2P overlay networks. In order to facilitate multimedia streaming and downloading service from servers, our caching scheme determines the appropriate availability of cached stream segments in a cache community, determines the appropriate peers for cache replacement, and performs availability-aware cache replacement. It achieves a small access latency for stream retrieval, stable bandwidth provisioning during retrieval session, and load balancing of clients' requests among peers.
In QoS-aware object replication, we formulate QoS-aware replication problem where each node specifies an upper bound on the time to access a given object. The problem is to minimize the number of replicas in order to satisfy these access deadlines. We show that this problem is intractable, since it is NP-complete. We present centralized as well as distributed solutions.