Research On Key Technologies Of Mobile Storage And Spatial Temproal Data Analysis In Cloud Environment

During the last decade, cloud computing has become an important technology foremerging applications. In cloud computing environments, there are many challengesrelated to large-scale data storage, management and analysis. In particular, how toaccess spatial temporal data in the cloud from mobile devices and how to analyze bigspatial temporal data using cloud computing are two challenging issues. First, with thegrowing popularity of mobile devices, they become an important source of the spatialtemporal data. It is challenging to provide an efficient and convenient way for mobiledevices to access cloud storage services. Second, it is common that data generated frommobile devices contain spatial and temporal information. How to efficiently analyzemassive spatio-temporal data is another important work of data processing in cloudenvironments. In this thesis, we focus on the topics of how to enable efficient access ofheterogeneous cloud storage services and how to provide analytical applications withscalable spatio-temporal data management technologies. The main contributions andinnovations of this thesis are as follows:(1) Proposes Wukong, a file system for transparently accessing heterogeneouscloud storage services. In the cloud environment, there is no uniform data accessinginterface for various storage services. As a result, an application is usually limited tousing specific interfaces for data accessing. This fact prevents spatial temporal datacollection applications from seamlessly accessing cloud storage services with differentinterfaces. Fortunately, the file system interface is supported by almost all applications.In Wukong, we seek to translate various cloud service interfaces into the uniform filesystem interface which will enable transparent access of heterogeneous cloud storageservices without modifying the applications. To the best of our knowledge, Wukong isthe first system for this purpose. Based on the bridge design pattern, Wukong uses theservice mashup technology to provide mobile applications that are only capable ofaccessing local files with the ability of accessing cloud storage services. Wukong usesan interface mapping method to support heterogeneous cloud services. Wukong alsoproposes several optimization strategies to improve the efficiency and the availability ofthe system. We have implemented a prototype and evaluated it in a systematic way.Experiments show that Wukong provides applications a transparent way to accessheterogeneous cloud services with high performance.(2) Proposes EaSync, a cross-device file synchronization service. Since peopleusually have more than one device nowadays, it is more and more common for usersswitching between multiple devices to complete their daily work. However, due to thedifferences of functionalities and computing power, different devices may needdifferent file sets. It is challenging to keep all devices synchronized and maintain multiple data views. In this thesis, we propose a file synchronization service acrossmultiple devices called EaSync. We design a dual-timestamp-based synchronizationprotocol to ensure data consistency. We also propose a tree-based approach to maintainmultiple data views during the file synchronization. We implement and evaluate theEaSync prototype system. As the result shown, EaSync outperforms othersynchronization systems like iFolder in the common operations.(3) Proposes TruthFinder, a ground truth exploration system for trajectory data. Toenable smart transportation, a large volume of vehicle GPS data has been collected inmany metropolitan-scale cities. The collected raw GPS data, however, suffers frommany problems such as low sampling rate, missing data, inaccurate data, and erroneousdata. Map matching is a commonly used technique that calibrates the raw GPS recordsby aligning them onto the corresponding road network. In practice, however, there islittle ground truth data to validate the calibration results. In this thesis, we proposeTruthFinder, an interactive map matching system for ground truth data exploration.TruthFinder incorporates traditional map matching algorithms and human intelligencein a unified manner. We also design a novel cost model that classifies and quantifieshuman operations to measure and optimize the interaction cost. TruthFinder is evaluatedin terms of accuracy and operation cost. The result shows that TruthFinder provides theusers an efficient way to explore ground truth from raw data.(4) Proposes parallel implementations of R-tree packing and DBScan algorithms.When dealing with spatio-temporal data, existing index packing methods and clusteranalysis algorithms cannot scale to very large datasets. In this thesis, we design andimplement a general framework for parallel Rtree packing using MapReduce. In thisframework, we pack each R-tree level from bottom up in parallel. This approachovercomes the memory and computation limitations of the sequential R-tree packingmethods. Besides, we propose MRDBScan, a parallel density-based clusteringalgorithm using MapReduce to address the scalability issue of classic DBScanalgorithms.In summary, the above key techniques will provide a strong technicalsupport for large-scale data storage, management and analysis in the cloud computingenvironment. We have evaluated the parallel algorithms on a12-nodes cluster.Experiments not only show that our approaches have a high performance but alsoconfirm the scalability of our approaches.With the studying of the above key techniques, we provide a strong technicalsupport for the area on data storage, management and analysis for large scale dataset.