Uma estratégia paralela e distribuída para assegurar a confidencialidade de dados armazenados em nuvem

Abstract

Nowadays, the storage of large amounts of confidential data in cloud servers is a common practice, since such strategy allows to reduce costs and also increases data availability. However, in cloud computing environment, data control is no longer owned by its legitimated user, becoming a storage service provider responsability. Such scenario gives rise to new challenges related to privacy, security and confidentiality. At this context, different solutions have been proposed for ensuring the confidentiality of the cloud stored data. In generall, such approaches are based on cryptography, data fragmentation or a combination of these two methodologies. Recently, a new approach, denoted QSM-EXTRACTION, has been proposed. The QSM-EXTRACTION strategy is based on the fragmentation of a digital file into fragments named information objects, on the decomposition of these objects through the extraction of some features and on the dispersion of these features in different cloud storage services. However, despite being developed for cloud computing environment, QSM-EXTRACTION method adopts a centralized execution approach, which may compromise the performance of the decomposition step. At the present work, we propose a paralell and distributed version of the QSM-EXTRACTION strategy, named pdQSM-EXTRACTION, which exploits the MapReduce paradigm aiming to provide a higher efficiency for the process of extracting features from information objects. The pdQSM-EXTRACTION approach has been implemented in Scala language programming, using Apache Spark framework. Several computing experiments and simulations have been performed aiming to evaluate the proposed approach. The obtained results, considering file sizes greater than or equal to 4GB, show that pdQSM-EXTRACTION strategy presents better performance than the one obtained by the QSM-EXTRACTION strategy, evaluated by computing the input time, defined as the total time spent to decompose a given file generating three other files containing the characteristics of quality, quantity and measurement. Thus, considering the processing of files whose sizes are greater than or equal to 4GB and the addition of one or more slave nodes by the pdQSM-EXTRACTION strategy, the ratio between the input time obtained by the pdQSM-EXTRACTION strategy and the input time obtained by the QSM-EXTRACTION strategy presented minimum and maximum values respectively of 53.57 % and 95.83 %. Therefore, we achieve to demonstrate the feasibility of pdQSM-EXTRACTION approach for applications involving large data volumes.