Real-time data backup in Geo-distributed data center networks against progressive disaster

In the progressive disaster scenario where multiple nodes and links are sequentially disrupted with different early warning times, real-time data backup is a critical concern to improve the survivability of a geo-distributed DCN (Geographically distributed Data Center Network). We consider this problem by assuming that different disaster-covered DCN nodes and links have different early warning times and different disaster-covered DCN nodes have distinct sets of backup nodes. Based on integer linear program (ILP), we first propose an optimal data backup scheme to maximize the total amount of data backup. This theoretically provides an upper bound on the total amount of data backup. Then, a heuristic (i.e., Amount-Based Algorithm) is developed to provide a time-efficient solution, in which an adjusting factor is introduced to deal with link bandwidth competitions among multiple disaster-covered DCN nodes. Furthermore, we give another way to increase the amount of data backup for DCN operators by adjusting the bandwidth distribution on the key link which is identified as the one that has the greatest impact on data backup. The extensive numerical results validate the proposed solutions under the progressive disaster scenario with different early warning times for different nodes and links.