An Adaptive Workload Prediction Strategy for Non-Gaussian Cloud Service Using ARMA Model with Higher Order Statistics

Abstract

With the increasing demand for cloud computing services, cloud providers are required to meet customers demand instantly by automatically scaling resources to users as needed. However, workloads are dynamically changing over time, and instantaneous resource allocation in the cloud is not possible due to the start-up time of the provisioning process. To address this problem, it is necessary for cloud providers to predict the future demand and provision resources in advance. In this paper, we propose an adaptive workload prediction model using Higher Order Statistics (HOS) with an Autoregressive Moving Average (ARMA) model. The proposed method makes use of HOS to perform a Gaussianity verification test of the workload. Based on the test results, different identification methods of the ARMA model are automatically assigned to predict the workload. In addition, the proposed method applies feedback from latest observed workloads to update the model on the run. Using real traces of requests to web servers, we conduct extensive experiments and show the efficiency of the proposed method.