Online adaptation models for resource usage prediction in cloud network
Shaifu Gupta, Dileep Aroor Dinesh · 2017
Cloud computing provides rapid on-demand access to shared services over the Internet. Elasticity is a key feature of cloud that allows the system to dynamically adapt to workload changes such that the resource allocation sufficiently models the resource usage as close as possible. It is a highly challenging job as a number of users enter and leave the cloud environment dynamically. Predicting the usage of different resources in advance helps the service providers in better capacity planning and satisfy the needs of users. In this paper, we analyze the multi-step ahead CPU usage predictions of different existing time series prediction models. To model the highly varying cloud workloads and counter the effect of error propagation in iterative multi-step ahead predictions we proposed and analyzed online adaptation of time series CPU usage prediction models. We also use fractional differencing to capture long range dependence in the data. We evaluate the proposed adaptive models on Google cluster trace.