No abstract

parte do requisitos para a obtenção do título de Mestre em Ciências -Área: Ciências de Computação e Matemática Computacional. USP -São Carlos Dezembro de 1998 Dedico este trabalho a meus pais, Evani e Emílio, que sempre me cercam de amizade e companheirismo. Sem o carinho, a confiança e o apoio a mim dedicados, nada disso seria possível. Agradecimentos À Profa. Dra. Regina Helena Carlucci Santana, pela paciência, pela orientação, tanto acadêmica como pessoal, e principalmente pela amizade. Ao Prof. Dr. Marcos José Santana, pelo incentivo, amizade e companheirismo. Aos amigos Paulo Sérgio, Márcio e Luciano pelo apoio e pelas contribuições feitas a este trabalho. Aos amigos que fiz durante o mestrado, sobretudo às pessoas do grupo de Sistemas Distribuídos e Programação Concorrente, pelos momentos descontraídos que passamos juntos. À todos os funcionários, que direta ou indiretamente, colaboraram para a realização deste trabalho. Um agradecimento especial à Adriana e à Sandra, da seção de bolsas, pela assistência e ao Rogério, do Centro de Informática de São Carlos -CISC, pela colaboração e paciência. Ao CNPQ e à FAPESP, pelo apoio financeiro. E principalmente, aos meus pais, que me apoiaram e me incentivaram nos momentos mais difíceis. Os resultados globais obtidos tornam possível verificar-se e demonstrar-se (para o ambiente particular considerado) que o PVM apresenta melhor desempenho em ambientes computacionais distribuídos e o MPI se comporta mais adequadamente em arquiteturas paralelas. AbstractThis work presents a performance evalualion of the portability platforms PVM and MPI when running on both a distributed computing system and a parallel architecturethe IBM SP2. The evaluation is performed on the implementation of a number of sorting parallel algorithms. Four implementalion of MPI and PVM are considered: the IBM MPI and the IBM PVMe (both running in the SP2); the MPICH and the PVM (both running in a distributed computing system).The parallel algorithms were developed and the results obtained are used to establish a comparison between the different environments and between the severa! sorting algorithms implemented. The sequential algorithms were also analysed allowing the speedup evalualion for the execufion in each environment.The global results obtained make possible to verify and to demonstrate (for the particular environment considered) that the PVM presents better performance in distributed computing environment and the MPI behaves more adequately in parallel architectures.7.

This paper proposes a novel approach for the implementation of simulation scenarios, providing a client entity for cloud computing systems. The client entity allows the creation of scenarios in which the client behavior has an influence on the simulation, making the results more realistic. The proposed client entity is based on several characteristics that affect the performance of a cloud computing system, including different modes of submission and their behavior when the waiting time between requests (think time) is considered. The proposed characterization of the client enables the sending of either individual requests or group of Web services to scenarios where the workload takes the form of bursts. The client entity is included in the CloudSim, a framework for modelling and simulation of cloud computing. Experimental results show the influence of the client behavior on the performance of the services executed in a cloud computing system.

Cloud Computing is the industry whose demand has been growing continuously since its appearance as a solution that offers different types of computing resources as a service over the Internet. The number of cloud computing providers grows into a run, while the end user is currently in the position of having many pricing options, distinct features and performance for the same required service. This work is inserted in the cloud computing task scheduling research field to hybrid cloud environments with service-oriented architecture (SOA), dynamic allocation and control of services and QoS requirements attendance. Therefore, it is proposed the QBroker Architecture, representing a cloud broker with trading features that implement the intermediation services, defined by the NIST Cloud Computing Reference Model. An experimental design was created in order to demonstrate compliance to the QoS requirement of maximum task execution time, the differentiation of services and dynamic allocation of services. The experimental results obtained by simulation with CloudSim prove that QBroker has the necessary requirements to provide QoS improvement in hybrid cloud computing environments based on SOA.