A workload prediction-based multi-VM provisioning mechanism in cloud computing

With the emerging of cloud computing, more and more enterprise organizations begin to migrate their applications to IaaS, which is a more flexible and cheaper alternative to traditional infrastructures. IaaS providers usually offer customers with resources in the form of VM and charge them in a time-based billing model. Meanwhile customers are allowed to dynamically apply for VM resources. However, highly dynamic workload makes customers difficultly determine how much capacity to provision. Furthermore, it is also a great challenge for customers to determine how to choose a VM provisioning scheme to match workload at a low cost. In this paper, we propose a workload prediction-based multi-VM provisioning mechanism to overcome these challenges, which contains an ARIMA workload predictor with dynamic error compensation (ARIMA-DEC) and a time-based billing aware multi-VM provisioning algorithm (TBAMP). The experimental results show that ARIMA-DEC predictor can obviously reduce SLA default rate and TBAMP algorithm can effectively save rental cost comparing to the existing algorithms.