A Novel Replica Placement Strategy for Data Center Network

With the rapid development of the cloud and data-intensive computing, many data center network are growing more large scale, and the number of servers is increasing at an exponential rate. As commodity-class PCs is used in the current DCN, failures caused by node failure, rack failures, link failures and routing failures become a very common phenomenon in the parallel processing of data blocks, tasks, and job scheduling. DCell is a novel network structure adapted to the DCN, which uses a recursively-defined structure to interconnect servers. Although DCell is fault tolerant, and addresses various failures, it has many questions. If the destination nodes are the failure, no matter what cannot route to them. The existing technology of DCell fail to guarantee durability and reliable for the data on the node. This paper proposes a novel idea that a replica is placed in the node of the layer by the fixed-point single-source shortest path, and there is at least one replica in each layer of the DCell. According to the proposed placement strategy of replicas, we suppose a mechanism to restore lost replica based on the neighboring nodes in the same path. Finally, we simulate and validate algorithm of placement and repair for replica. Experiments show that our algorithm has obvious effect to improve the reliability of data.