A memory stores a log that includes first, second, and third datasets each containing server information collected at different times. The processor forms vectors from the datasets, calculates distance vectors between them and uses the distance vectors and pairwise angles between distance vectors to predict a dataset for a first future time. When the first future time arrives, the processor collects the server information, stores it in a fourth dataset, and calculates a variance between that dataset and the predicted dataset. Using the variance, distance vectors, and angles, the processor determines a second predicted dataset corresponding to a second future time, from which it determines a server usage prediction. If the server usage prediction is below a decommission threshold, the processor decommissions the server. If the server usage prediction is above a capacity threshold, the processor generates a report advising an administrator to upgrade or remove projects from the server.