Labrador: towards fair and auditable data sharing in cloud computing with long-term privacy

Xiaojie Guo, Jin Li, Zheli Liu, Yu Wei, Xiao Zhang, Changyu Dong

Abstract: Data are one of the most important sources of power that drives the world today. However, aggregating data is not an easy task with increasing legal regulations and concerns from users about their data privacy, and therefore incentives might be needed to encourage data sharing. In this paper, we present Labrador (LB), a system to handle the above problems. Our result demonstrates long-term privacy that reveals only an analytic result to the data analyst. An analytic task is delegated to clouds, which holds users’ homomorphically encrypted data. We develop a lightweight verifiable blind decryption technique over the linearly homomorphic encryption scheme to verify the final result. Thus, its verifiability and blindness rely on over-determined and under-determined systems, respectively. To support incentives in data sharing, we leverage smart contract to realize binding contracts between mutually distrusted parties. In the game theory model with a non-collusion assumption, Labrador is secure against any rational adversary. Our evaluation demonstrates that the computational overhead for the data analyst and the data owner is insignificant (i.e., only a few seconds and milliseconds, respectively).