Heitor Faria, Rommel Carvalho and Priscila Solis, University of Brasilia (UnB), Brazil

Backup software information is a potential source for data mining: not only the unstructured stored data from all other backed-up servers, but also backup jobs metadata, which is stored in a formerly known catalog database. Data mining this database, in special, could be used in order to improve backup quality, automation, reliability, predict bottlenecks, identify risks, failure trends, and provide specific needed report information that could not be fetched from closed format property stock property backup software database. Ignoring this data mining project might be costly, with lots of unnecessary human intervention, uncoordinated work and pitfalls, such as having backup service disruption, because of insufficient planning. The specific goal of this practical paper is using Knowledge Discovery in Database Time Series, Stochastic Models and R scripts in order to predict backup storage data growth. This project could not be done with traditional closed format proprietary solutions, since it is generally impossible to read their database data from third party software because of vendor lock-in deliberate overshadow. Nevertheless, it is very feasible with Bacula: the current third most popular backup software worldwide, and open source. This paper is focused on the backup storage demand prediction problem, using the most popular prediction algorithms. Among them, Holt-Winters Model had the highest success rate for the tested data sets.

Backup, Catalog, Data Mining, Forecast, R, Storage, Prediction, ARIMA, Holt-Winters