Privacy-preserving workflow scheduling in geo-distributed data centers

Due to the increasing volume of data to be analyzed and the need for global collaborations, many scientific applications have been deployed in a geo-distributed manner. Scientific workflows provide a good model for running and managing geo-distributed scientific data analytics. However, due to the multi-level data privacy requirements in geo-distributed data centers (DCs), as well as the costly and heterogeneous inter-DC network performance, executing scientific workflows efficiently in such a geo-distributed environment is not easy. In this paper, we propose a privacy-preserving workflow scheduling algorithm named PPPS, which aims at minimizing the inter-DC data transfer time for workflows while satisfying data privacy requirements. We compare PPPS with five state-of-the-art workflow scheduling algorithms using Windows Azure cloud performance traces and real scientific workflows. Experimental results show that PPPS can greatly reduce the workflow execution time compared to the other algorithms by up to 93% while satisfying complicated data privacy constraints.