有限存储分布式社交网络中数据的动态划分和复制

社交网络的庞大数据需求分布式存储,多个用户的数据分散存储在各个存储和计算节点上可以保持并行性和冗余性。如何在有限的分布式存储空间内高性能存储和访问用户数据具有现实意义。在当前的社交网络系统中,用户数据之间的读写操作会导致大量跨存储节点的远程访问。减少节点间的远程访问可以降低网络负载和访问延迟,提高用户体验。提出一种基于用户交互行为的动态划分复制算法,利用用户之间的朋友关系和评论行为描述社交网络的结构,周期性划分复制用户数据,从而提高本地访问率,降低网络负载。通过真实数据集验证,该算法相比随机划分和复制算法能够大大提升本地访问率,降低访问延迟。

Dynamic Partition and Replication Algorithm for Storage Capacity Bound Distributed Social Network

Large scale social networks require distributed storage systems. The data of users are distributed on multiple storage and computing nodes to provide parallelism and redundancy. How to store and access user data effectively within limited distributed storage has practical significance. However, there are a large number of remote accesses due to the community structure of social network. Decreasing remote accesses would reduce the network load and access latency, and improve user experience. This paper proposes a dynamic partitioning and replication algorithm based on user interactions, which describes community structure of social network by user relations and comments. It partitions and replicates user data periodically, in order to improve local access ratio and reduce network communication load. Through experiment on a real-world dataset, this paper proves that compared with random par- titioning and replication, this algorithm could improve local access ratio greatly, and finally reduce access latency.