Research On Replica Optimization In Data-intensive Computing

With the dramatically soaring of the Internet users and the increasing popularity of the broadband network, data plays an increasing important role in people's production and life, and Internet service gradually became a kind of service that centers on mass data processing, its Quality of Service(QoS) heavily depends on the ability to process data provided. Data-intensive computing emerges as required.Replica management is a key technology in data-intensive application, it can not only improve the reliability and availability of data in data-intensive environment, still can effectively reduce the network delay and increase load balance in network. Therefore, to optimize the replica management is an effective way to improve the Quality of Service (QoS) of data-intensive applications.Based on the studies on replica management in data-intensive computing environment, and the analysis of the research status at home and abroad, this thesis deeply researches replica selection and replica replacement in replica management, and propose corresponding optimization strategy. The thesis mainly includes two aspects as following:(1)Optimization for replica selection:According to the commercial characteristics in cloud computing, based on Weighted Set Covering Problem (WSCP), this thesis proposes a cost-aware replica selection approach, such that it can select low cost replicas and reduce data transfer time simultaneously to improve efficiency of applications, and also extend the CloudSim from two aspects including file transferring and dynamic bandwidth, and verify our algorithm in the thesis.(2)Optimization for replica replacement:Based on the studies of traditional replica replacement algorithms, for the inabiligy of LRU in weak localigy and the weaknesses of traditional replacement strategies, enlightened by the LIRS(Low Inter-reference Recency) cache replacement proposed by Xiaodong Zhang, who is a chair professor in Ohio State University, this thesis implements an LIRS-based replica replacement strategy, extends the OptorSim based on LIRS, and comprehensively compares this strategy with other strategies(LRU, MRU, LFU, MFU) by use of different scheduling algorithms and file access patterns respectively.To sum up, based on the studies and analysis of replica management mechanism, the thesis proposes optimization strategies for replica selection and replica replacement respectively, and simulates our stategies in corresponding simulator. The experiments results show that the strategies proposed in this thesis has certain advantages in some aspects.