A Completion Time-Based Flow Scheduling for Inter-Data Center Traffic Optimization

With the rise of cloud computing service, traffic in inter-data center wide area networks (inter-DC WANs) has been growing rapidly. Inter-data center traffic contains multitudinous data transfers that require to be accomplished within certain time periods or deadlines. The different completion time of the transfer tasks has distinct impacts on the performance of the whole system. However, very little work takes the completion time into account while guarantees the hard deadlines. Besides, most previous works arrange transfers exclusively, which does not take fairness into account. In this paper, we introduce a completion time-based model, which takes multiple deadlines and their impacts into consideration to profile a more accurate characterization of the data transfers. We propose a novel mechanism to achieve the goals of both maximizing utility and fairness. We also develop an inter-DC WAN emulation tool which enables a single commodity server to emulate arbitrary topology and dynamic configuration on the WAN links. Our evaluation demonstrates that our algorithm gains 110%-250% utility than current deadline focus mechanisms, while achieves 12% higher throughput. Fairness is improved over four times at best by adjusting the fairness coefficient.