Abstract: Univariate time series forecasting can be used to dynamically tune the resource allocation for databases. It is vital that relational databases have adequate storage for archive logs since lack of space can cause the database to hang, while overallocation can reduce the efficiency of utilization. Most of the time, storage is allocated for the peak usage and kept for the duration of its lifecycle. This paper presents a conceptual model that uses predictive analysis to dynamically scale the storage allocation for archive logs generated by databases. The framework presented in this paper does exploratory data analysis on archive logs data, compares the accuracy of various statistical models such as autoregressive integrated moving average (ARIMA), Holt damped trend, Holt linear trend, Mean, Naïve and Seasonal Naive models, suggests the best model suited for each database, and provides forecast of storage usage. These predictions can be used as input in other automated systems to automatically provision the storage or repurpose unused storage as needed.
Authors: Gazanfur A Mohammed (Cisco Systems, USA)