Abstract
Data centers have powerful processing capabilities, large storage, and backup power supplies to store the users' data and provide tools that enable users to process their data. Availability is one of the main issues in data center. Recently, the raising of always-on online services has made the availability of data center more important. However, the virtual machines (VMs) running in sites inside a data center are under the risk of hardware failure. This made the challenge to make the service always-on. Backing up VM is one way to improve the availability of data center. Live VMs have to periodically update itself to remote sites to keep backups up to date. When a live VM is down due to hardware failure, then one of the backup VMs in the other sites will be turned on to keep the service going on. Due to the fact that the used network bandwidths of VMs are seldom to reach the requested network bandwidths of VMs at the same time, we can minimize backup cost by increasing the network bandwidth oversubscription. In this thesis, we introduce a novel minimum non-metric costcost VM backup problem with network bandwidth oversubscription. Next, we prove the introduced problem is NP-hard. Due to the fact underlying the network bandwidth-oversubscription model is that statistically few users will attempt to fully utilize their requested network bandwidth simultaneously, we propose a heuristic adaptive bandwidth-oversubscription algorithm to backup live VMs. We conduct simulations, using the semisynthetic data to compare the proposed algorithm with some greedy-based heuristics. The simulation results show that the proposed algorithm outperforms these heuristics.