Research On Load Scheduling And Replica Management Of Clustered VoD System

Pursuing performance-price ratio and reducing TCO are the most important goal of VoD service with small and medium scale. The Bluegrid clustered VoD system based on Open source software and cheap PC implements a solution for VoD service with the feature of simple deployment, convenient maintenance and has high applied cost.Efficient data management mechanism and load scheduling strategy are the foundation to improve throughput of VoD system. Distributed storage strategy used in large scale VoD service improves the system performance but also brings the problem of complexity in data management and poor scalability. The most existing load scheduling strategies currently are centralized in distributed storage model, and in shared storage model, there is few strategies offered for dispatching requests according to real-time load.For meeting the requirement of small and medium scale VoD service, the system uses RAID for centralized storage and combines local disk of real servers to implement a kind of 2-level hierarchical storage architecture. By analyzing the distribution of VoD requests, combining distribution with statistical information, adopting linear regression to forecast the popularity of movies, calculating the place value, an adaptive replica management mechanism has been implemented to resolve problem of replicas placement in local disk. Consequently, in result, the mechanism reduces the I/O pressure on RAID and also improves the response time.Based on the 2-level hierarchical storage architecture, an application-specific scheduling algorithm called Greatest Capability of Weighted Load First has been proposed. The algorithm can be used for dynamic load scheduling considering node-state, node performance and weighted load factor product. The algorithm can be able to improve the load balancing effectively.The experimental results show that adaptive replica management mechanism resolves replica placement in 2-level hierarchical storage architecture and improves load balancing. GCWLF can guarantee the linear growth of system performance as well as the good scalability.