CRUPA: A container resource utilization prediction algorithm for auto-scaling based on time series analysis
Yang Meng, Ruonan Rao, Xin Zhang, Pei Hong · 2016
Computing resource utilization of applications may vary over time and inappropriate static resource provision would cause resource wastage or performance loss. Auto scaling based on instant demand is a simple solution but it also introduces latency of scaling. An accurate resource demand prediction algorithm is able to eliminate the side effects of scaling. To address this issue, we present CRUPA, a resource utilization prediction algorithm based on a time series analysis model (ARIMA) combined with Docker container technique. We also design a comparison experiment to evaluate the proposed algorithm and its average forecast error is only 6.5% in the short term, which is much lower than the most common model based on threshold (16.9%) on the same dataset. The result shows that CRUPA not only has high prediction accuracy but also scales the resource well.