Study On Regional Cloud Storage Based On Two-layer Judgement And Average Access Cost

Cloud storage is a novel storing method based on cloud architecture. With thegrowing of Mobile Internet and Big Data applications, the demand of cloud storagehas been growing rapidly. Due to cross network bandwidth bottleneck, it is commonfor cross-geography regional clouds to use distributed storage and resourcelocalization technologies. Aiming at network and geography complexity of regionalcloud systems, determining how to increase data availability, reduce response andaccess time, and reduce network resource consumption has become a new subject ofstudies on regional clouds.In order to analyze regional cloud systems, this paper proposed a structuralmodel of regional cloud system and the concept of access cost, and analyzed possiblefactors that may influence access costs. In the analysis, we focus on availablebandwidth, distance between sites and availability of data files and blocks; and byquantizing these factors, we deduced a mathematical model for access costs andaverage access costs. Based on the model, we proposed an average-access-cost-driven,adaptive, two-layered dynamic replication strategy (two-layer average access cost,T-AAC). The strategy calculates the popularity of a specific data file according to sitestatus, regional datacenter status and data access history; combining file popularitywith file availability, it gives out replica creation timing for both within and acrossregional data centers. T-AAC also defines a replica deletion strategy within, whichcan be divided into two stages: first use least frequently used (LFU) algorithm tomake preliminary selection, then use average cost algorithm to execute final deletion.T-AAC is an adaptive dynamic replication strategy, it can adaptively create and deletereplicas, optimize replica distribution at each regional data center and site, reduce dataaccess time and bandwidth consumption.We extended OptorSim simulation platform to perform experiments. Resultsshow that, compared to traditional least frequently used (LFU) and least recent used (LRU) algorithms, T-AAC could reduce average job time, network bandwidthconsumption and cross regional data center file visits, system performances thusimproved.