Time Series Based Bandwidth Allocation Strategy in Cloud Datacenter

摘要

Network bandwidth is a critical resource for tenants in cloud datacenter. Early researches assume that tenants are aware of bandwidth demands of the virtual machines, and a fixed bandwidth capacity can be satisfying. However, bandwidth demands of different virtual machines are complex and time-varying and it is insufficient to re-allocate the resource when congestion or low utilization occurs. Our proposal aims to design and implement a bandwidth allocation system embedded in cloud platform with technology of software-defined network (SDN). The allocation system analyses statistic data periodically taken from running virtual machines and forecasts bandwidth utilization of each VM for the next period. After that, it generates the bandwidth allocation strategy based on the prediction and pre-allocates bandwidth to virtual machines. The experiments show that our auto pre-allocation system can effectively improve the network performance of cloud datacenter, both in utilization and capacity.