Submitted by Automa??o e Estat?stica (sst@bczm.ufrn.br) on 2017-11-01T21:17:51Z No. of bitstreams: 1 LeoMoreiraSilva_DISSERT.pdf: 4274864 bytes, checksum: ba5d7987907f47bb678e8a7cb507ff0c (MD5) / Approved for entry into archive by Arlan Eloi Leite Silva (eloihistoriador@yahoo.com.br) on 2017-11-07T22:32:56Z (GMT) No. of bitstreams: 1 LeoMoreiraSilva_DISSERT.pdf: 4274864 bytes, checksum: ba5d7987907f47bb678e8a7cb507ff0c (MD5) / Made available in DSpace on 2017-11-07T22:32:56Z (GMT). No. of bitstreams: 1 LeoMoreiraSilva_DISSERT.pdf: 4274864 bytes, checksum: ba5d7987907f47bb678e8a7cb507ff0c (MD5) Previous issue date: 2017-07-26 / A manuten??o e a evolu??o dos sistemas de software podem trazer v?rias mudan?as de c?digo que podem potencialmente reduzir sua qualidade e aumentar sua complexidade. Um atributo de qualidade cr?tico que ? afetado ao longo do tempo ? o desempenho do sistema. Assim, sem o devido acompanhamento, esse atributo de qualidade pode deixar de ser atendido adequadamente. A ?rea de visualiza??o de software prop?e o uso de t?cnicas cujo objetivo ? melhorar o entendimento do software e tornar mais produtivo o seu processo de desenvolvimento. Neste contexto, este trabalho apresenta o PerfMiner Visualizer - uma ferramenta para visualizar e analisar desvios de desempenho em evolu??es subsequentes de um sistema de software. Atrav?s de visualiza??es de grafos de chamadas e sumariza??o de cen?rios, a ferramenta permite que desenvolvedores e arquitetos possam identificar cen?rios e m?todos que tiveram varia??es no seu desempenho, inclusive as potenciais causas desses desvios atrav?s dos commits. O trabalho tamb?m apresenta um estudo emp?rico que avalia o uso da ferramenta aplicando-a em 10 vers?es de evolu??o de 2 sistemas open source de dom?nios diferentes e submetendo question?rios online para obter feedback dos seus desenvolvedores e arquitetos. Os resultados do estudo conduzido trazem evid?ncias preliminares da efic?cia das visualiza??es providas pela ferramenta em compara??o com dados tabulares. Al?m disso, o algoritmo de supress?o de n?s da visualiza??o do grafo de chamadas foi capaz de reduzir entre 73,77% e 99,83% a quantidade de n?s a serem exibidos para o usu?rio, permitindo que ele possa identificar mais facilmente as poss?veis causas das varia??es. / The maintenance and evolution of software systems can bring several code changes that can potentially reduce their quality and increase their complexity. One critical quality attribute that be affected over time is the system performance. Thus, without due monitoring, the performance quality attribute may no longer be adequately addressed. The software visualization area proposes the use of techniques whose objective is to improve the understanding of the software and to make its development process more productive. In this context, this work presents PerfMiner Visualizer - a tool to visualize and analyze the performance deviations from subsequent evolutions of a software system. Through call graph and scenario summarization visualizations, the tool allows developers and architects identifying scenarios and methods that have variations in their performance, including the potential causes of such deviations through commits. This work also presents an empirical study that evaluates the use of the tool by applying it to 10 evolutionary versions of 2 open source systems from different domains and by submitting online questionnaires to obtain feedback from their developers and architects. The results of the conducted study bring preliminary evidence of the effectiveness of visualizations provided by the tool compared to tabular data. In addition, the nodes suppression algorithm of the call graph visualization was able to reduce between 73.77% and 99.83% the number of nodes to be displayed to the user, allowing him to be able to identify more easily the possible causes of variations.