The Design And Implementation Of File System Based On PaaS Cloud Storage Service

With the widespread using of computers and the trend of mobile Internet, usersenjoy the convenience and experience on seamless online services in distinctscenarios and devices, and have to be confronted with the data management issues aswell. Further, a multitude of innovative applications, such as coordinative working inoffice hinge on the development of data sharing and synchronization. To some degree,the blooming of variable network disk services evidences that the demand existsindeed currently. Both individuals and enterprises are seeking the approach toboosting the effectiveness of data management.In this paper, we propose the design and implementation of a novel file systembase on the PaaS (Platform as a Service) Cloud storage service in order to pursue abetter pattern of data sharing.We first investigate the basic primitives, usage model, and billing mechanisms ofthe PaaS storage service, and then analyze the effects of operations for data block in aconventional file system. According to the features of the Cloud service, the mainprinciple is to reduce the frequency of accessing the remote server by optimizing thelayout of concrete objects in the Cloud and the cache replacement policy so as toreduce the latency of accessing WAN data and lower the monetary cost. In particular,three metrics that depict the correlation of blocks and files as well as the hotspot areselected to determine the organization of the storage object and the cache evictionmethod. Next, with this core, we design a series of components to support the targetfile system, including the strengthened file system interface in the user space, theembedded database for metadata, the garbage cleaner, and the compression andencryption utilities.Then, we implement a prototype system that incorporates a hierarchy andinterface-oriented programing, which corresponds to flexibility that each layer ofconcrete module that can be replaced. Besides the proposed strategy of dataorganization and caching that is based on the k-d tree algorithm, we also compare oursystem with some related work. Furthermore, because the data consistency issue fordifferent service providers is masked by some techniques, additional features, such asdeduplication and snapshot, increase the availabi lity and performance. Finally, toevaluate the system, we choose variable simulating and real world workloads under the network environment that crosses the continent. The evaluation results show thatour work is superior to the existing studies on cost, e xecution time, garbage cleaningeffectiveness and cache hit ratio.