Submitted by Rafael Santana (rafael.silvasantana@ufpe.br) on 2017-05-04T17:58:30Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) tese_rubens_digital_biblioteca_08092016.pdf: 4506490 bytes, checksum: 251226257a6b659a6ae047e659147a8a (MD5) / Made available in DSpace on 2017-05-04T17:58:30Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) tese_rubens_digital_biblioteca_08092016.pdf: 4506490 bytes, checksum: 251226257a6b659a6ae047e659147a8a (MD5) Previous issue date: 2016-03-01 / CAPES / Cloud computing paradigm is able to reduce costs of acquisition and maintenance of computer systems, and enables the balanced management of resources according to the demand. Hierarchical and composite analytical models are suitable for describing performance and dependability of cloud computing systems in a concise manner, dealing with the huge number of components which constitute such kind of system. That approach uses distinct sub-models for each system level and the measures obtained in each sub-model are integrated to compute the measures for the whole system. Identification of bottlenecks in hierarchical models might be difficult yet, due to the large number of parameters and their distribution among distinct modeling levels and formalisms. This thesis proposes methods for evaluation and detection of bottlenecks of cloud computing systems. The methodology is based on hierarchical modeling and parametric sensitivity analysis techniques tailored for such a scenario. This research introduces methods to build unified sensitivity rankings when distinct modeling formalisms are combined. These methods are embedded in the Mercury software tool, providing an automated sensitivity analysis framework for supporting the process. Distinct case studies helped in testing the methodology, encompassing hardware and software aspects of cloud systems, from basic infrastructure level to applications that are hosted in private clouds. The case studies showed that the proposed approach is helpful for guiding cloud systems designers and administrators in the decision-making process, especially for tune-up and architectural improvements. It is possible to employ the methodology through an optimization algorithm proposed here, called Sensitive GRASP. This algorithm aims at optimizing performance and dependability of computing systems that cannot stand the exploration of all architectural and configuration possibilities to find the best quality of service. This is especially useful for cloud-hosted services and their complex underlying infrastructures. / O paradigma de computação em nuvem é capaz de reduzir os custos de aquisição e manutenção de sistemas computacionais e permitir uma gestão equilibrada dos recursos de acordo com a demanda. Modelos analíticos hierárquicos e compostos são adequados para descrever de forma concisa o desempenho e a confiabilidade de sistemas de computação em nuvem, lidando com o grande número de componentes que constituem esse tipo de sistema. Esta abordagem usa sub-modelos distintos para cada nível do sistema e as medidas obtidas em cada sub-modelo são usadas para calcular as métricas desejadas para o sistema como um todo. A identificação de gargalos em modelos hierárquicos pode ser difícil, no entanto, devido ao grande número de parâmetros e sua distribuição entre os distintos formalismos e níveis de modelagem. Esta tese propõe métodos para a avaliação e detecção de gargalos de sistemas de computação em nuvem. A abordagem baseia-se na modelagem hierárquica e técnicas de análise de sensibilidade paramétrica adaptadas para tal cenário. Esta pesquisa apresenta métodos para construir rankings unificados de sensibilidade quando formalismos de modelagem distintos são combinados. Estes métodos são incorporados no software Mercury, fornecendo uma estrutura automatizada de apoio ao processo. Uma metodologia de suporte a essa abordagem foi proposta e testada ao longo de estudos de casos distintos, abrangendo aspectos de hardware e software de sistemas IaaS (Infraestrutura como um serviço), desde o nível de infraestrutura básica até os aplicativos hospedados em nuvens privadas. Os estudos de caso mostraram que a abordagem proposta é útil para orientar os projetistas e administradores de infraestruturas de nuvem no processo de tomada de decisões, especialmente para ajustes eventuais e melhorias arquiteturais. A metodologia também pode ser aplicada por meio de um algoritmo de otimização proposto aqui, chamado Sensitive GRASP. Este algoritmo tem o objetivo de otimizar o desempenho e a confiabilidade de sistemas em cenários onde não é possível explorar todas as possibilidades arquiteturais e de configuração para encontrar a melhor qualidade de serviço. Isto é especialmente útil para os serviços hospedados na nuvem e suas complexas

Cloud datacenters comprise hundreds or thousands of disparate application services, each having stringent performance and availability requirements, sharing a finite set of heterogeneous hardware and software resources. The implication of such complex environment is that the occurrence of performance problems, such as slow application response and unplanned downtimes, has become a norm rather than exception resulting in decreased revenue, damaged reputation, and huge human-effort in diagnosis. Though causes can be as varied as application issues (e.g. bugs), machine-level failures (e.g. faulty server), and operator errors (e.g. mis-configurations), recent studies have attributed capacity-related issues, such as resource shortage and contention, as the cause of most performance problems on the Internet today. As cloud datacenters become increasingly autonomous there is need for automated performance diagnosis systems that can adapt their operation to reflect the changing workload and topology in the infrastructure. In particular, such systems should be able to detect anomalous performance events, uncover manifestations of capacity bottlenecks, localize actual root-cause(s), and possibly suggest or actuate corrections. This thesis investigates approaches for diagnosing performance problems in cloud infrastructures. We present the outcome of an extensive survey of existing research contributions addressing performance diagnosis in diverse systems domains. We also present models and algorithms for detecting anomalies in real-time application performance and identification of anomalous datacenter resources based on operational metrics and spatial dependency across datacenter components. Empirical evaluations of our approaches shows how they can be used to improve end-user experience, service assurance and support root-cause analysis. / Cloud Control (C0590801)