Explorando modelos preditivos para prever características e comportamentos de aplicações.

Abstract

In a globalized and competitive market, companies need processes that are focused on the efficient use of their resources. In relation to computational costs, they have been benefiting in recent decades with the adoption of cloud computing, which has promoted significant savings for companies. However, this scenario has stabilized and new challenges arise in the management of these resources. To manage the efficient use of computational resources, software and algorithms that use predictive models to predict behaviors and react to adapt their infrastructure to the demand of use in real time are emerging. This paper aims to explore predictive machine learning techniques to predict application behaviors based on historical memory consumption, request, and latency in order to determine how to better manage these resources. The analysis was divided into pre-processing and data cleaning, then the data was subjected to algorithm processing that detects behaviors embedded in the data and a prediction was generated that was used to estimate future application behaviors. To help detect application behaviors from historical consumption data and estimate their behavior, the paper discusses different techniques related to time series prediction. These estimates can be used to perform infrastructure augmentation or downsizing, based on inactivity or resource overuse behaviors.