IEEE Project Abstract

service level agreement (SLA) management is one of the key issues in cloud computing. The primary goal of a service provider is to minimize the risk of service violations, as these results in penalties in terms of both money and a decrease in trustworthiness. To avoid SLA violations, the service provider needs to predict the likelihood of violation for each SLO and its measurable characteristics (QoS parameters)and take immediate action to avoid violations occurring. There are several approaches discussed in the literature to predict service violation; however, none of these explores how a change in control parameters and the freshness of data impact prediction accuracy and result in the effective management of an SLAof the cloud service provider. The contribution of this paper is two-fold. First, we analyzed the accuracy of six widely used prediction algorithms simple exponential smoothing, simple moving average, weighted moving average, Holt Winter double exponential smoothing, extrapolation, and the auto regressive integrated moving average by varying their individual control parameters. Each of the approaches is compared to 10 different data sets at different time intervals between 5 min and 4 weeks. Second, we analyzed the prediction accuracy of the simple exponential smoothing method by considering the freshness of a data;i.e., how the accuracy varies in the initial time period of prediction compared to later ones. To achieve this, we divided the cloud QoS data set into sets of input values that range from 100 to 500 intervals insets of 1 100, 1 200, 1 300, 1 400, and 1 500. From the analysis, we observed that different prediction methods behave differently based on the control parameter and the nature of the data set. The analysis helps service providers choose a suitable prediction method with optimal control parameters so that they can obtain accurate prediction results to manage SLA intelligently and avoid violation penalties.