Research On Fault Tolerant Storage And Data Access Optimization In Data Center Networks

Data center, of which performance affects the cloud computing, is the key part of cloud computing technology. As clouding computing has be en penetrated into our daily life, the data stored in the data center has increased rapidly, which leads to the fact that the requirements of fault-tolerance of distributed storage system have to be improved. In recent ten years, introducing the network coding into the data fault-tolerant technology is one of the hot research topic. After the introduction of network coding in distributed system, the questions of placement strategy and data access in data center can be expanded to new discussions. This topic is an optimization research on these two issues.The first part of this topic is designing the optimized placement strategy, which is based on the fault-tolerant storage. There are three types of data center network architecture style, switch-oriented, server-centric and hybrid network structure. The server-centric structure is more symmetrical on the distance between any two servers. Then, the BCube data center network, one popular server-centric network structure, is chosed in this topic. On the optimized plcement strategy proposed in this topic, there are three f actors taken into consideration: the data access efficiency, storage node loads, and the expectations of recovery time of failure nodes. Integrating the above three factors, the standard of choosing the distributed storage nodes has been maken. Due to the node loads have been more balanced, and recovery time for failure nodes has been shorter, the structure of storage nodes in accordance with the standard has a higher fault tolerance.In the second part of this topic, according to the characteristics of cod ed system, a new scheduling policy of user requests has been proposed. In the coded system, several data stored in several node s can be decoded into complete original data. Hence, users' requests can be serverd in any node. Then, the number of service windows in queue model has increased. Undoubtedly, the avrage delay will be reduced. The scheduling algorithm proposed in this part, which has increased the number of queues in the buffer, has managed the users' requests in the buffer by the method of setting different priorities of queues. So there is no need to memory the served child requests, the server will not be idle while child requests in waiting will not assign their case on the occurrence. Through such management, the user access efficiency in coding system was optimized.Finally, simulate the above two algorithms, to verify whether the placement strategy and scheduling policy is optimized.