Higher order statistics based method for workload prediction in the cloud using ARMA model

Abstract

One of the main features that distinguishes the cloud computing from the other computing models is the ability to dynamically scale resources for users as needed. However, allocating resources in the cloud is not instantaneous; it takes several minutes for a virtual instance to be initialized. The obvious solution to this issue is to predict the future need of computing resources and allocate them before being requested. In this paper, we present a workload forecasting model based on the Autoregressive Moving Average (ARMA) model for both Gaussian and non-Gaussian processes. In the proposed method, the technique of Higher Order Statistics (HOS) is used in conjunction with the ARMA model to identify the appropriate forecasting model in case of Gaussian and non-Gaussian processes. Results of experiments show the efficiency of the proposed method compared to the traditional ARMA model.