Abstract: In order to improve performance while satisfy the availability of the system in cloud storage. And reduce the network bandwidth consumption in the process of replica synchronization. An adaptive mechanism of replica consistency based on the file heat is proposed in this paper. This mechanism selects the corresponding consistency strategies at runtime according to the file heat. Strong consistency service is used in the circumstance that the file heat is high. Oppositely, eventual consistency is used for the file with low heat to save valuable network resources. The file heat is calculated by file access time sequence and LRFU algorithm, the file heat value indicates the future access and is consistent with the actual file access situation. Evaluation results show that the proposed mechanism reduce the network bandwidth consumption while satisfy the availability of the system, balance the availability and performance of the system.

Key words: cloud storage, replica, consistency, adaptive, heat, multi-replica

摘要： 在云存储系统中，为了保证系统可用性的同时尽可能提高系统性能，降低副本同步过程中网络带宽等资源的消耗。提出了基于文件热度的副本自适应一致性方法。该方法根据文件的热度不同，动态调整副本同步的一致性策略，对于热度高的文件在副本同步过程中采用强一致性，对于热度低的文件在副本同步过程中采用最终一致性。在计算文件热度时，考虑了文件访问的时间序列并结合LRFU算法，该热度值表示了文件将来可能的访问情况，并与文件的实际访问情况一致。实验表明该方法在保证系统可用性的同时显著降低了网络带宽资源的消耗，有效地平衡了系统的可用性和性能。

关键词: 云存储, 副本, 一致性, 自适应, 热度, 多副本