PANP-GM: A periodic adaptive neighbor workload prediction model based on grey forecasting for cloud resource provisioning

Abstract

Cloud computing platforms provide on-demand service to meet users’ need by adding or removing cloud resources dynamically. The cloud resource provisioning is often based on the feedback model, which causes time delay and resource wasters. Workload prediction methods can make the resource provisioning more instantaneous and reduce resource and power consumption, to meet service level objectives (SLOs) and improve quality of service (QoS) of cloud platform. In this paper, we propose a periodic adaptive neighbor workload prediction model based on grey forecasting (PANP-GM) for cloud resource provisioning. Firstly, the model analyzes the growth rate and evaluates the periodicity of workload. Secondly, this model uses the growth rate of previous neighbor periodicity to predicate the trend of upcoming workload. To adapt to dynamic changes and emergencies, the grey forecasting model is applied for automatic error correction and improving prediction accuracy. Experimental results demonstrate that PANP-GM can achieve better resource prediction accuracy than basic and general approaches. Furthermore, this model can effectively improve the QoS of cloud platform and reduce SLO violations.