BenchXtend: uma ferramenta para medir a elasticidade de sistemas de banco de dados em nuvem

• Autor(a) principal: Rodrigo FÃlix de Almeida
• Data de Publicação: 2013
• Tipo de documento: Dissertação
• Idioma: eng
• Título da fonte: Biblioteca Digital de Teses e Dissertações da UFC
• Texto Completo: http://www.teses.ufc.br/tde_busca/arquivo.php?codArquivo=10911
• topic: Elasticidade; Bancos de dados; Benchmarking; Elasticity; Databases; Cloud; CIENCIA DA COMPUTACAO

In recent years, cloud computing has attracted attention from industry and academic world, becoming increasingly common to find cases of cloud adoption by companies and research institutions in the literature. Since the majority of cloud applications are data-driven, database management systems powering these applications are critical components in the application stack. Many novel database systems have emerged to fulfill new requirements of high-scalable cloud applications. Those systems have remarkable differences when compared to traditional relational databases. Moreover, since elasticity is a key feature in cloud computing and it is a differential of this computing paradigm, novel database systems must also provide elasticity. Altogether with the emergence of these new systems, the need of evaluating them comes up. Traditional benchmark tools for database systems are not sufficient to analyze some specificities of these systems in a cloud. Thus, new benchmark tools are required to properly evaluate such cloud systems and also to measure how elastic they are. Before actually benchmarking and measuring elasticity of cloud database systems, it becomes necessary to define a model with elasticity metrics that makes sense both for consumers and providers. In this work we present BenchXtend, a tool, that extends Yahoo! Cloud Serving Benchmark (YCSB), to benchmark cloud database systems and to measure elasticity of such systems. As part of this work, we propose a model with metrics from consumer and provider perspectives to measure elasticity. Finally, we evaluated our solution by performing experiments and we verified that our tool could properly vary the load during execution, as expected, and that our elasticity model could capture the elasticity differences between the studied scenarios.