Research On Data Caching Based On Data Correlation In Mobile Cloud Service

With the popularity of mobile devices,mobile cloud computing has received widespread attention and has gradually become a major form of cloud services.Data caching can migrate shared data to some advantage nodes to serve the mobile users' data requests by mining the spatio-temporal trajectories of mobile user access,which can not only greatly improve the service quality of cloud service and reduce network latency and load,but also dramatically reduce service cost of mobile cloud services.In this paper,we study the data caching problem where multiple data items are cached,transferred,replicated as well as deleted in a mobile cloud environment to serve a pre-defined sequence of requests.There are two distinct inherent features in this problem,which are not available to traditional scenario.First,the data caching problem in mobile cloud services is typically cost-oriented,instead of capacity-oriented,because the storage capacity can be virtually viewed as infinite as long as user can afford it.Thus,the cost,not the hit ratio in traditional scenarios,is the key concern in this paper.Second,data items are often correlated,not independent as considered in traditional cases,and their accesses in mobile cloud services are in general trajectory-based.With these features in mind,this paper proposes a two-phase caching algorithm for multiple shared data items in mobile cloud services.By leveraging the access trajectories of requests,we first investigate the correlation among data items to determine whether or not two or more data items could be packed together to be transfered,and then propose a hybrid strategy with the name DP_Greedy,to combine the dynamic programming algorithm in previous works and a greedy algorithm to effectively cache these shared data items to serve a pre-defined sequence of requests.The essence of this algorithm is based on a general observation that two or more packed data items to serve requests jointly are usually more cost effective than each individual one.Under homogeneous cost model,we prove the proposed algorithm is at most 2/? times worse than the optimal one in terms of total service cost of a pre-defined request sequence,where ? is the discount factor we defined.The proposed algorithm is evaluated by experiments,and the results show DP_Greedy not only performs excellently,but also is more in line with actual situation.