A framework for privacy-aware computing on hybrid clouds with mixed-sensitivity data

Xiangqiang Xu

Cloud computing has significantly increased the computation/storage capacity for regular users, which leads to the popularity of transplanting large-scale computations, most likely big data applications, to clouds. In fact, a large fraction of big data applications contain sensitive data. However, high-level security cannot always be guaranteed by cloud service providers, thus sensitive information can be easily exposed during the process of data processing. Hybrid clouds provide potentials for handling data separately based on their sensitivity, harnessing the heterogeneous architecture. In this thesis, we design and implement a privacy-aware framework to address data privacy challenges by supporting sensitive data segregation on hybrid clouds. On the one hand, to guarantee data privacy, sensitive data are tagged and retained on the private cloud, so that data sensitivity can be segregated from the public cloud. On the other hand, to ensure performance, we introduce a simple optimization model, which assigns certain amount of non-sensitive data to the public cloud for processing. To achieve our goal of guaranteeing data privacy as well as improving performance, we propose three tagging mechanisms to handle data with mixed-sensitivity, which includes a coarse-grained file level tagging, a fine-grained line level tagging, and a dynamic line level tagging for processing data generated on-the-fly. Specifically, static data with mixed-sensitivity can be handled by using either the file level tagging or the static line level tagging mechanism. Dynamically generated data with mixed-sensitivity can be handled by using the dynamic line level tagging mechanism, in which data are processed dynamically, and tags can be added or removed at run-time. We demonstrate the effectiveness of these three mechanisms by evaluating them using a big data application. Our experimental results show that the privacy-aware framework successfully enables data sensitivity protection while providing good performance. Our framework shows good scalability when the data sensitivity level increases. In addition, our results demonstrate that our dynamic line level tagging has a stable and reliable performance for processing near-real-time data.

A framework for privacy-aware computing on hybrid clouds with mixed-sensitivity data

Files and links (1)

Abstract

Metrics

Details