Avaliação da validade externa da técnica de análise diferencial para detecção de envelhecimento de software: um estudo confirmatório com replicação

• Ano de defesa: 2017
• Autor(a) principal: Sena, Guilherme Otávio de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Uberlândia; Brasil; Programa de Pós-graduação em Ciência da Computação
• Palavras-chave em Português: Computação; Comportamento - Analise experimental; Software - Verificação; Sistemas de memória de computadores; Envelhecimento de Software; Vazamento de Memória; Detecção de Anomalias; Análise Diferencial; Software Aging; Memory Leak; Anomaly Detection; Differential Analysis; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufu.br/handle/123456789/19851; http://doi.org/10.14393/ufu.di.2017.503

Software systems running continuously for a period of time often confront software aging. This phenomenon is related to the increase of the failure rate as the system executes. Recently, a study introduced a technique for aging detection based on differential software analysis that, through experiments under synthetic workloads with focus on memory leakage, proved superior to other approaches used in SAR. The differential analysis can distinguish between the natural behavior of aging behavior when comparing (under experiments) two versions of the same system: target version (with aging) and base version (without aging). This master's study evaluated the external validity of this approach to verify if the previous findings also applied to real applications and loads. For this purpose, 4 widely known real-world applications with memory leak bugs were selected. The activation patterns of the bugs were studieds in order to incorporate them into the representativeness characterization of the workload scenarios used. Subsequently, for each application, experimental replications were performed on the target and base versions considering the planned workload scenarios. In each replication, the RSS and HUS indicators were monitored, each composing a different time series. Then, in order to reduce the dissimilarity effects between the series, a mean time series was estimated by the DTW method for each set of replications. Finally, the mean time series of each indicator were processed through a combination of statistical techniques of trend detection and CEP, generating divergence charts for the anomalies identification. The divergence charts allow a fair comparison of the leak detection performance of each technique/indicator combination. The results showed that, unlike those previously findings with synthetic workloads, all combinations were able to detect memory leak efficiently, with no false-negatives and few false-positives rates. In addition, the trend detection techniques, in particular Hodrick Prescott (HP), were better than those of CEP. Again, the HUS indicator was superior to RSS, determining HP/HUS as the best overall combination to detect memory leakage.