Dynamic characteristics of user interactivity make supporting VCR-like operations in peer-to-peer (P2P) Video on Demand (VoD) streaming systems very challenging. Recently, the prediction-based prefetching of hot segments scheme has emerged as a promising approach to improve user Quality of Experience. However, this prediction model uses a centralized server to collect and analyze the large volumes of user viewing logs for predicting user VCR behavior. This log server can easily become bottleneck in terms of data exchange and processing. In this paper, we propose a novel distributed load balancing and VCR-aware two-tier P2P VoD System (DLCA). DLCA relies on a two-tier architecture. In the low tier, the common nodes form a classic gossip-based unstructured network for normal data distribution. On the top layer, a portion of strong nodes establish a structured DHT network for VCR-related information analysis and publish. By employing a pattern mining algorithm, each strong mode maintains a prefetching routing table, which can effectively assist common nodes prefetching segments for VCR-like interactivity during playback. Simulation results show how DLCA outperforms a state of the art centralised method in terms of performance.
展开▼
