Coordenação de Aperfeiçoamento de Pessoal de Nível Superior / Computer grids emerged in the 90 s with the goal of using geographically dispersed computers for high performance computing. Through grids, computational power of a supercomputer can be reached in a simple, efficient and inexpensive way. Such benefits led to highlights in researchs of computer grids. Recently, appeared on market graphics adapter cards whose computational power overcomes, and by a wide margin, even the most modern processors commonly used. This led to researchs that resulted in programming techniques relatively easy to learn and did simplify application programming for these processors. These techniques effectively introduced the processors in the business of high performance computing. The use of these techniques gave rise to General Purpose computing on Graphic Processing Units (GPGPU). Grids applications are generally programmed through a grid computing framework. TUXUR is one of those frameworks and is under development by Master s Program Graduate at the Federal University of Santa Maria. This dissertation discusses the development of a TUXUR s foreseen feature. Such feature allows the computer grid managed by TUXUR to enjoy the benefits of GPGPU applications, particularly regarding to the best use of the nodes s hardware that comprises it. The immediate impact of this synergy is the significant increase in grid computational capacity without adding new computers. The findings of the evaluation highlights the importance of using GPGPU tasks that take advantage of this programming technique, even when performed in a grid. / Grades de computadores surgiram na década de 90 com o objetivo de utilizar computadores geograficamente dispersos para computação de alto desempenho. Através destas grades, pode-se chegar ao poder computacional de um supercomputador de uma forma simples, eficiente e barata. Tais benefícios fizeram com que pesquisas em grades de computadores obtivessem destaques no ramo da computação. Recentemente, surgiram no mercado placas adaptadoras gráficas cujo poder computacional supera, e com larga vantagem, mesmo os mais modernos processadores de uso geral. Isso deu origem a pesquisas que resultaram em técnicas de programação relativamente fáceis de aprender e que simplificam a programação de aplicações para estes processadores. Estas técnicas efetivamente introduziram estes processadores no ramo de computação de alto desempenho. A utilização destas técnicas deu origem à programação de propósito geral em unidades de processamento gráfico (General-Purpose computation on Graphical Processor Units-GPGPU). Aplicações de grades são geralmente programadas através de um framework de computação em grade. TUXUR é um destes frameworks e encontra-se em desenvolvimento por mestrandos do Programa de Pós-graduação em Informática da Universidade Federal de Santa Maria. Este trabalho aborda o desenvolvimento de uma funcionalidade prevista no TUXUR. Tal funcionalidade permite que a grade de computadores gerenciada por TUXUR usufrua dos benefícios de aplicações GPGPU, sobretudo no que diz respeito à melhor utilização do hardware dos nós que a compõem. O reflexo imediato desta sinergia é o aumento significativo da capacidade computacional da grade sem o acréscimo de novos computadores. Os resultados encontrados na avaliação evidenciam a importância do uso de GPGPU nas tarefas que se beneficiam desta técnica de programação, mesmo quando executadas em uma grade.

Many of the grid compute resources (CPU cycles, network bandwidths, computing power, processor times, and software) exist as non-storable commodities, which we call grid compute commodities (gcc) and are distributed geographically across organizations. These organizations have dissimilar resource compositions and usage policies, which makes pricing grid resources and guaranteeing their availability a challenge. Several initiatives (Globus, Legion, Nimrod/G) have developed various frameworks for grid resource management. However, there has been a very little effort in pricing the resources. In this thesis, we propose financial option based model for pricing grid resources by devising three research threads: pricing the gcc as a problem of real option, modeling gcc spot price using a discrete time approach, and addressing uncertainty constraints in the provision of Quality of Service (QoS) using fuzzy logic. We used GridSim, a simulation tool for resource usage in a Grid to experiment and test our model. To further consolidate our model and validate our results, we analyzed usage traces from six real grids from across the world for which we priced a set of resources. We designed a Price Variant Function (PVF) in our model, which is a fuzzy value and its application attracts more patronage to a grid that has more resources to offer and also redirect patronage from a grid that is very busy to another grid. Our experimental results show that the application of the PVF has helped achieve equilibrium between users satisfaction measured as QoS and recovery of the infrastructure investment made by the providers. In the absence of pricing benchmarks, we setup Commodity Base Prices (CBP) and then integrated our PVF and CBP with GridSim to price grid compute resources. In summary, this thesis provides the design of a model to price grid compute resources using financial options theory. The model achieves mutual benefit for users and providers in the grid environment. The mutual benefit is expressed in terms of QoS to the users and recovery of investments on the grid infrastructure for the providers. This thesis has opened up many different opportunities for further research especially in the era of enterprise computing with clouds.

Many of the grid compute resources (CPU cycles, network bandwidths, computing power, processor times, and software) exist as non-storable commodities, which we call grid compute commodities (gcc) and are distributed geographically across organizations. These organizations have dissimilar resource compositions and usage policies, which makes pricing grid resources and guaranteeing their availability a challenge. Several initiatives (Globus, Legion, Nimrod/G) have developed various frameworks for grid resource management. However, there has been a very little effort in pricing the resources. In this thesis, we propose financial option based model for pricing grid resources by devising three research threads: pricing the gcc as a problem of real option, modeling gcc spot price using a discrete time approach, and addressing uncertainty constraints in the provision of Quality of Service (QoS) using fuzzy logic. We used GridSim, a simulation tool for resource usage in a Grid to experiment and test our model. To further consolidate our model and validate our results, we analyzed usage traces from six real grids from across the world for which we priced a set of resources. We designed a Price Variant Function (PVF) in our model, which is a fuzzy value and its application attracts more patronage to a grid that has more resources to offer and also redirect patronage from a grid that is very busy to another grid. Our experimental results show that the application of the PVF has helped achieve equilibrium between users satisfaction measured as QoS and recovery of the infrastructure investment made by the providers. In the absence of pricing benchmarks, we setup Commodity Base Prices (CBP) and then integrated our PVF and CBP with GridSim to price grid compute resources. In summary, this thesis provides the design of a model to price grid compute resources using financial options theory. The model achieves mutual benefit for users and providers in the grid environment. The mutual benefit is expressed in terms of QoS to the users and recovery of investments on the grid infrastructure for the providers. This thesis has opened up many different opportunities for further research especially in the era of enterprise computing with clouds.