Study on Replica Strategy Based on Access Pattern Mining in Smart City Cloud Storage System

摘要

The replica strategy in traditional distributed file system, which creates a copy mainly from the perspective of internal resources while changes in external demand are ignored. However, this strategy is not suitable for deployment in a service-based, resource-rich internal storage “smart city” in cloud storage center. This paper proposes a replica strategy, which combines data security (the minimum amount of copies) together with service needs (best copy volume). The strategy predicts file popularity based on access pattern mining algorithms. What’s more, the number of copies of the cloud adjusts itself dynamically according to the popularity of file and system resources. Mining algorithm is based on the analysis of the characteristics of spatio-temporal data in smart cities. The algorithm first maps the historical user access request to the spatio-temporal attribute domain. Then according to the geographical area grid and association rules, the correlation analysis and evolution rule identification of access requests are carried out in the domain of spatio-temporal attributes. Finally dig out the user access mode and predict the user’s access request, calculate the file popularity according to the request. The simulation results show that the popularity of the file calculated by the access pattern mining algorithm in this paper is simple and efficient, and the prediction accuracy of the popularity can reach 84%. The dynamic replica mechanism based on popularity has a significant advantage in coping with sudden large-scale concurrent accesses. Meanwhile, compared with the conventional dynamic replicas based on access frequency, the proposed strategy consumes less storage resources.