Uploading multiply deferrable big data to the cloud platform using cost-effective online algorithms

摘要

Cloud computing consists of processing big data and provides convenient, on-demand network access to a shared pool of configurable computing resources. Cloud data center costs have become a hot topic in recent years. To minimize bandwidth costs, a better solution for uploading multiply deferrable big data to a cloud computing platform for processing using a MapReduce framework was studied. The multiply deferrable big data, which have its own delay window sizes, are produced by local cloud users, and the bandwidth charging model in this paper is the Max contract pricing scheme adopted by Internet service providers (ISPs). A basic single-ISP case was analyzed. We then extended the study to the cloud scene. The Multi-Heuristic Smoothing Algorithm for the single case was designed, and we proved that the worst-case competitive ratio of the Multi-Heuristic Smoothing Algorithm falls between 2(1 － (1 － l/D_(max))~(D_(max)) and 2, where D_(max) is the maximum delay window size. In addition, the Multi-Dynamic Self-Adaption Algorithm (MDSA) was designed to optimize the cloud scene based on the Multi-Heuristic Smoothing Algorithm. The simulation experiments demonstrated that the total cost was reduced by 12% when the Multi-Dynamic Self-Adaption Algorithm was adopted.