Section: New Results
PerForecast: A Tool to Forecast the Evolution of Time Series for Capacity Planning.
Participants: C. Leverger, R. Marguerie, V. Lemaire, T. Guyet, S. Malinowski
The work published in  presents PerForecast, a tool for automatic capacity planning. The tool relies on univariate temporal data and automatically configured predictive models. The aim is to anticipate capacity problems in the infrastructure of Orange in order to ensure the delivery of services to customers. For example, PerForecast can predict the overhead of a server at the earliest possible stage, so that new machines can be ordered before the deterioration of the service in question. The purchase procedures being long and costly, the earlier they are done, the better the quality of service.