A Cooperative Placement Method for Machine Learning Workflows and Meteorological Big Data Security Protection in Cloud Computing

Abstract

Cloud computing has proven to be a powerful paradigm in both academia and industry. A variety of meteorological applications using machine learning modeled as the workflows and meteorological big data have been accommodated in the meteorological cloud infrastructure. However, it still faces challenges to guarantee the execution enciency of the meteorological machine-learning workflows and avoid the privacy leakage of the datasets in a semi-trusted cloud. To tackle this challenge, a collaborative placement method (CPM) and a two-factor-based protection mechanism for machine-learning workflows and big data security protection is proposed. Technically, fat-tree topology is leveraged to institute the meteorological cloud infrastructure. Then, the non-dominated sorting differential evolution (NSDE) technique is employed to realize joint optimization of data access time, energy efficiency and load balance. In terms of security protection, the proposed mechanism allows data owners (DOs) to send encrypted data to users through meteorological cloud server (MCS). The DOs are required to formulate access policy and perform ciphertext-policy attribute-based encryption (CP-ABE) on data. In order to decrypt, the users need to possess two factors that a secret key and a security device (e.g., a sensor card in meteorological applications). The ciphertext can be decrypted if and only if the user gathers the secret key and the security device at the same time. Eventually, the experiment evaluates the performance of CPM.