A Method to Cloud Computing Resources Requirement Prediction on SaaS Application

Abstract

Nowadays, it is more important than ever before to accurately predict the cloud computing resource requirement on Software as Service (SaaS) applications. However, the extant methods for requirement prediction are largely designed from the perspective of cloud resource providers, which just results in the insufficient scalability of the models. To solve the insufficiency, a novel method described as Extended Coupled Hidden Markov Model (ECHMM) is proposed in this paper. Based on the historical monitoring data, this paper pioneers to take into account two factors of software evolution and resources correlations in the prediction. Combined with the characteristics of SaaS, it is thus possible for the construction of the relationship between multiple observation states and hidden states. Baum-Welch algorithm is used for the training of the model parameters and the obtaining of the optimal solution. And on a typical SaaS application platform, a cloud computing resource prediction experiment is carried out. The experiment shows that a fuller range of cloud computing resources: the prediction result of the model is closer to the real value. The prediction errors of CPU., memory and network resources are reduced by 2%, 1.7% and 4.5% respectively. And the model can catch an “abnormal” state and get it corrected in time for the increase of scalability.