Bandwidth Reservation for Data Transfers through Multiple Disjoint Paths of Dynamic HPNs

Abstract

Bandwidth reservation was initially designed to provide QoS for real-time multimedia applications, and has been widely used since because of its proved high efficiency and effectiveness. It was recently started being used for big data transfers in high-performance networks from the data generation center to remotely located collaborating sites for data storage, mining and analysis. Most existing work on big data transfer using bandwidth reservation service focuses on addressing various data transfer problems on one or two network paths. In this paper, we study three important problems regarding data transfers using bandwidth reservation through multiple (node-and edge-)disjoint paths in dynamic high-performance networks: (i) achieve the earliest completion time for a given data transfer request through multiple variable disjoint paths, (ii) achieve the earliest completion time for a given data transfer request through multiple fixed disjoint paths with fixed bandwidths, and (iii) achieve the minimum number of fixed disjoint paths with fixed bandwidths to finish the given data transfer by its deadline. We prove all of these problems to be NP-complete and then propose one heuristic algorithm for each. We compare the heuristic algorithms with existing scheduling algorithms, and conduct extensive simulations. The simulation results show that our proposed heuristic algorithms have much better overall scheduling performance.