END-TO-END PERFORMANCE ANALYSIS OF A RESOURCE ALLOCATION SERVICE

• Autor(a) principal: Mateus, Marcos Miguel Teixeira Pereira
• Data de Publicação: 2022
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/155752
• topic: DataMiner; Data Mining; Classification; Regression; Machine Learning; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

This dissertation is centered around the monitoring and control platform of the company Skyline Communications, the DataMiner. This platform has a specific module called SRM (Service and Resource Management). One of SRM’s many features is the capacity to make an advance reservation (booking) of resources of the client’s network. When a booking is created, there is a time interval/delay between the moment that a booking is requested and the moment that all necessary configurations for this booking actually start to be made at the Resource level. This delay is called SyncTime. The SyncTime is affected by the dynamics of the network (e.g, a sudden increase in the number of bookings made, at a given time). In order to guarantee the maximum possible quality of service to the client and ensure that the network dynamics will have minimal impact in the real-time delivery of the desired content, the SRM module must be able to estimate if a booking can be done in an acceptable SyncTime value. Given this problem, the main goal of this dissertation is to develop a machine learning based estimation/classification module that is capable of, based on the temporal state of the SRM module, make a prediction or classification of the SyncTime. Two approaches were considered: Classify the SyncTime based on classes or estimate the value of it. In order to test both approaches, we implemented several traditional machine learning methods as well as, several neural networks. Both approaches were tested using a dataset collected from a DataMiner cluster composed of three DataMiner agents using software developed in this dissertation. In order to collect the dataset, we considered several setups that captured the cluster in different network conditions. By comparing both approaches, the results suggested that classifying the predicted SyncTime using a classification model and classifying the predicted SyncTime of a estimation model are both equally good options. Furthermore, based on the results of all implementations, a prototype application was also developed. This application was fully developed in Python and it uses a Multilayer Perceptron in order to do the classification of the SyncTime of a booking, based on several inputs given by the user.