Short-term forecasting models for automated data backup system: segmented regression analysis

Abstract

The Information and Communication Technology (ICT) becomes a critical area to business success; organizations need to adopt additional measures to ensure the availability of their services. However, such services are often not planned, analyzed and monitored, which impacts the assurance quality to customers. The backup is the service addressed in this study, with the object of study of the automated data backup systems in operation at the Federal University of Itajuba - Brazil. The main objective of this research was to present a logical sequence of steps to obtain short-term forecast models that estimate the point at which each recording media reaches its storage capacity limit. The input data was collected in the metadata generated by the backup system, with 2 years data window. For the implementation of the models, the simple univariate linear regression technique was employed in conjunction, in some cases, with the simple segmented linear regression. In order to discover the breakpoint, a targeted approach to residual analysis was applied. The results obtained by the iterative implementation of the proposed algorithm showed adherence to the characteristics of the analyzed series, with accuracy measures, regression significance, normality residual through control charts, model adjustment, among others. As a result, an algorithm was developed for integration into automated backup systems using the methodology described in this study.