Design And Implementation Of Distributed Media Cluster Systems

The paper proposes wide-area distributed media clusters and studies their designs and implementations. Compared to traditional media clusters, these kinds of clusters have more complex structure, but they can surmount limitations of traditional media clusters to provide resolutions with good quality, large capacity and high cost performance.We suggest a model of distributed cluster systems, called Cluster Group Tree (CG tree), and provide its major operation algorithms that cover heartbeat delivery within CG tree, CG tree image merging, and CG tree adjusting. The model can efficiently reduce the cost of intercommunication between different networks, and can automatically adjust CG tree to ensure high availability when cluster faults occur.We use MPICH to implement a distributed media cluster prototype, and discuss advantages and drawbacks that use this traditional parallel scientific computation tool to implement management software of distributed media clusters. Our practice demonstrates that the use of MPICH as a rapid prototype development tool can avoid repeated calling many complicated low level communication primitives, keep away from details concerned with running environment system, and save us a lot of development time. The result of our experiment also indicates that this distributed media cluster has quite good performance.