| The produce-study-research cooperation among the organizations, including enterprises,universities, and institutes etc., can effectively accelerate the industrialization,commercialization and globalization of scientific and technological achievements. TechnologyInnovation Platform (TIP) just provides a platform to such cooperation, which can help allparticipators to overcome the geographic obstacles and to cooperate at any time, integrateeducation, technology and business into a produce-study-research organism. A key to constructTIP is data storage. On this platform, data from various regions ought to be stored. Besidesthis, data should be stored in a cost-availability-effective way. Therefore, how to ensure theability of cross-region and cost-availability-effective storage, which is the foundation of theactivities on the platform, is a vital problem need to be addressed. Thus, the aim of thisdissertation is to provide an effective storage scheme for TIP, and design a correspondingfault-tolerance mechanism so as to meet the storage demand of TIP.Considering the demand for cross-regional storage ability, a cloud storage frameworknamed Hadoop Distributed File System (HDFS) is introduced to construct the storage systemof TIP. However, the previous version of HDFS adopts whole-copy redundant fault-tolerancemechanism. This mechanism ensures the high availability of system through copying andstoring the duplications respectively, which also increases the redundancy of the system andleads to the growth in cost. To avoid this situation, erasure codes are introduced into the designof new redundant fault-tolerance mechanism. In this new mechanism, data are encoded intoseveral check codes which can decode the lost data, so that the availability of system isenhanced. Compared with the whole-copy redundant fault-tolerance mechanism, this newmechanism can effectively reduce the redundancy and lower the storage cost.Additionally, to lower the cost while availability is enhanced, the cost-availabilityoptimization is included in the new redundant fault-tolerance mechanism. Firstly, abio-objective optimization model for cost and availability is established, and then aredundancy optimization algorithm based on Self-Organization Feature Map (SOFM) isdeveloped to search optimal solutions. Compared with other multi-objective optimizationapproaches, this algorithm is more adaptive to be applied under the cloud storage environmentdue to its parallelism, distribution and low complexity. In addition, in order to further improvethe performance on convergence and solutions, a parallel redundancy optimization algorithmis developed by introducing training modular and competition modular based on the previousalgorithm. Finally, the proposed redundant optimization algorithms and redundant fault-tolerancemechanism are tested. The experimental results prove that, compared with othermulti-objective optimization approaches, the new algorithm proposed by this dissertationperforms better on convergence and solutions. In addition, the results also prove that,compared with whole-copy redundant fault-tolerance mechanism, the new mechanismproposed by this dissertation not only ensures high availability, but also effectively reduces theredundancy and storage cost.The TIP storage system based on HDFS framework can effectively meet the platform’sdemands for cross-region. Meanwhile, the corresponding redundant fault-tolerance mechanismis proved to ensure high availability while the storage cost is reduced obviously, which canmeet practical demand for the data storage of TIP. Besides this, this dissertation also doessome work on the multi-objective optimization algorithms applied under the cloud storageenvironment through studying the redundancy optimization algorithms based on SOFM. |
PDF Full Text Request