On Periodic Point-to-Point Bandwidth Scheduling in High-Performance Networks

Today's network services and applications are widely distributed across multiple data centers around the world, and require timely transfer of big data between data centers to ensure the quality and reliability of services. Network infrastructure that connects data centers is expensive and such big data transfer would incur high cost, a large portion of which is due to the consumption of bandwidth resources. This paper aims to investigate the problem of minimizing data transfer cost in high-performance networks. We construct rigorous models for link bandwidth and link cost between data centers and formulate a Cost Minimization Bandwidth Scheduling Problem (CMBSP). Considering the differences in time and space of transmission cost, we focus on the problem of bandwidth redundancy con-sumption and cost waste caused by discontinuous time step transmission. We formulate CMBSP as an optimization problem of path selection and link bandwidth allocation, and design a cost timing allocation algorithm (CTAA), which allows users to transfer as much data as possible over inter-data center links and minimize the bandwidth cost. Simulation-based experiments in real-life network topologies such as ESnet and NSFNET show that CTAA outperforms state-of-the-art solutions.