Indiana University-Purdue University Indianapolis (IUPUI) / Performance models are useful as mathematical models to reason about the behavior of different computer systems while running various applications. In this thesis, we aim to provide two distinct performance models: one for distributed-memory high performance computing systems with network communication, and one for deep neural networks. Our main goal for the first model is insight and simplicity, while for the second we aim for accuracy in prediction. The first model is generalized for networked multi-core computer systems, while the second is specific to deep neural networks on a shared-memory system.

Productivity and quality of a manufacturing process are seen as important factors for short and long term success of a company. Measuring productivity and quality provides the means to assess performance trends of the process, providing a powerful management tool for decision making. Data Envelopment Analysis is a tool that can be used to measure productivity and quality performance. Productivity and quality performance were measured using Data Envelopment Analysis for a newspaper preprint insertion process. Hybrid performance models combining productivity and quality performance were also evaluated using Data Envelopment Analysis. Finally, the possible underlying relationship between productivity and quality was investigated. The major findings of this research are: * Productivity and quality performance trends obtained from the study provide realistic assessment of performance of the newspaper preprint insertion process. * The trends of the hybrid performance models were not able to accurately assess the performance of the newspaper preprint insertion process. * There does not seem to be a link between productivity and quality in the newspaper preprint insertion process for the time period studied. / Master of Science

No description available.

Este trabalho apresenta o desenvolvimento e a adaptação de modelos de previsão de desempenho aos fatores que condicionam as atividades de manutenção e reabilitação dos pavimentos nas cidades brasileiras de médio porte. Faz-se uma revisão bibliográfica sobre Sistemas de Gerência de Pavimentos e modelos de desempenho. Realiza-se um estudo de caso com seções selecionadas na rede pavimentada urbana de Araraquara-SP, exemplo de cidade brasileira de· médio porte. São desenvolvidos dois tipos de modelos: modelos empíricos, baseados em análises estatísticas de dados de pavimentos; e modelos subjetivos, com o desenvolvimento de matrizes de probabilidades de transição a partir de métodos formais de entrevistas com especialistas. Analisa-se a influência dos seguintes fatores sobre o desempenho de pavimentos urbanos: ano de construção, classificação funcional, atividades de manutenção e reabilitação, tráfego e estrutura do pavimento. Verifica-se uma significativa influência das atividades de manutenção e reabilitação. Nota-se, também, a necessidade de calibração periódica dos modelos de desempenho. Sugere-se a manutenção de arquivos de dados com informações sobre os fatores que afetam o desempenho dos pavimentos urbanos e sobre avaliações frequentes da condição dos pavimentos. / This work presents the development of performance prediction models, considering the factors that govern the programming of pavement maintenance and rehabilitation in Brazilian medium-sized cities. A bibliographic revision about Pavement Management Systems and performance models is pointed out. It is shown a case study with sections selected in the urban pavement network of Araraquara-SP, example of Brazilian medium-sized city. Two types of performance prediction models are developed: empirical models, based on statistics analyses of pavement data; and subjective models, based on the development of transition probability matrices from expert opinions. It is investigated the influence of the following factors on the performance of urban pavements: year of construction, functional category, maintenance and rehabilitation activities, traffic, and pavement structure. An expressive influence of the factor maintenance and rehabilitation activities is verified. It is also noticed the necessity of periodic calibration of the performance models. Maintenance of data files with information about the factors that act on the performance of urban pavements, and about frequent evaluations of the pavement condition is suggested.

Este trabalho apresenta o desenvolvimento e a adaptação de modelos de previsão de desempenho aos fatores que condicionam as atividades de manutenção e reabilitação dos pavimentos nas cidades brasileiras de médio porte. Faz-se uma revisão bibliográfica sobre Sistemas de Gerência de Pavimentos e modelos de desempenho. Realiza-se um estudo de caso com seções selecionadas na rede pavimentada urbana de Araraquara-SP, exemplo de cidade brasileira de· médio porte. São desenvolvidos dois tipos de modelos: modelos empíricos, baseados em análises estatísticas de dados de pavimentos; e modelos subjetivos, com o desenvolvimento de matrizes de probabilidades de transição a partir de métodos formais de entrevistas com especialistas. Analisa-se a influência dos seguintes fatores sobre o desempenho de pavimentos urbanos: ano de construção, classificação funcional, atividades de manutenção e reabilitação, tráfego e estrutura do pavimento. Verifica-se uma significativa influência das atividades de manutenção e reabilitação. Nota-se, também, a necessidade de calibração periódica dos modelos de desempenho. Sugere-se a manutenção de arquivos de dados com informações sobre os fatores que afetam o desempenho dos pavimentos urbanos e sobre avaliações frequentes da condição dos pavimentos. / This work presents the development of performance prediction models, considering the factors that govern the programming of pavement maintenance and rehabilitation in Brazilian medium-sized cities. A bibliographic revision about Pavement Management Systems and performance models is pointed out. It is shown a case study with sections selected in the urban pavement network of Araraquara-SP, example of Brazilian medium-sized city. Two types of performance prediction models are developed: empirical models, based on statistics analyses of pavement data; and subjective models, based on the development of transition probability matrices from expert opinions. It is investigated the influence of the following factors on the performance of urban pavements: year of construction, functional category, maintenance and rehabilitation activities, traffic, and pavement structure. An expressive influence of the factor maintenance and rehabilitation activities is verified. It is also noticed the necessity of periodic calibration of the performance models. Maintenance of data files with information about the factors that act on the performance of urban pavements, and about frequent evaluations of the pavement condition is suggested.

Parallel and distributed programming constitutes a highly promising approach to improving the performance of many applications. However, in comparison to sequential programming, many new problems arise in all phases of the development cycle of this kind of applications. For example, in the analysis phase of parallel/distributed programs, the programmer has to decompose the problem (data and/or code) to find the concurrency of the algorithm. In the design phase, the programmer has to be aware of the communication and synchronization conditions between tasks. In the implementation phase, the programmer has to learn how to use specific communication libraries and runtime environments but also to find a way of debugging programs. Finally, to obtain the best performance, the programmer has to tune the application by using monitoring tools, which collect information about the application's behavior. Tuning can be a very difficult task because it can be difficult to relate the information gathered by the monitor to the application's source code. Moreover, tuning can be even more difficult for those applications that change their behavior dynamically because, in this case, a problem might happen or not depending on the execution conditions.It can be seen that these issues require a high degree of expertise, which prevents the more widespread use of this kind of solution. One of the best ways to solve these problems would be to develop, as has been done in sequential programming, tools to support the analysis, design, coding, and tuning of parallel/distributed applications. In the particular case of performance analysis and/or tuning, it is important to note that the best way of analyzing and tuning parallel/distributed applications depends on some of their behavioral characteristics. If the application to be tuned behaves in a regular way then a static analysis (predictive or trace based) would be enough to find the application's performance bottlenecks and to indicate what should be done to overcome them. However, if the application changes its behavior from execution to execution or even dynamically changes its behavior in a single execution then the static analysis cannot offer efficient solutions for avoiding performance bottlenecks. In this case, dynamic monitoring and tuning techniques should be used instead. However, in dynamic monitoring and tuning, decisions must be taken efficiently, which means that the application's performance analysis outcome must be accurate and punctual in order to effectively tackle problems; at the same time, intrusion on the application must be minimized because the instrumentation inserted in the application in order to monitor and tune it alters its behavior and could introduce performance problems that were not there before the instrumentation. This is more difficult to achieve if there is no information about the structure and behavior of the application; therefore, blind automatic dynamic tuning approaches have limited success, whereas cooperative dynamic tuning approaches can cope with more complex problems at the cost of asking for user collaboration. We have proposed a third approach. If a programming tool, based on the use of skeletons or frameworks, has been used in the development of the application then much information about the structure and behavior of the application is available and a performance model associated to the structure of the application can be defined for use by the dynamic tuning tool. The resulting tuning tool should produce the outcome of a collaborative one while behaving like an automatic one from the point of view of the application developer.

Performance models are useful as mathematical models to reason about the behavior of different computer systems while running various applications. In this thesis, we aim to provide two distinct performance models: one for distributed-<br>memory high performance computing systems with network communication, and one for deep neural networks. Our main goal for the first model is insight and simplicity, while for the second we aim for accuracy in prediction. The first model is generalized for networked multi-core computer systems, while the second is specific to deep neural networks on a shared-memory system.<br>

Performance modeling is widely used to efficiently and rapidly assess the ability of multiprocessor architectures to effectively execute a given algorithm. In a typical design environment, VHD L performance models of hardware components are interconnected to form structural models of the target multiprocessor architectures. Algorithm features are described in application specific tools. Other automated tools partition the software among the various processors. Performance models evaluate the system performance. Since several iterations may be needed before a suitable configuration is obtained, a set of tools that directly interfaces the VHDL performance models to the algorithm partitioning tools will significantly reduce the time and effort needed to prepare the various models. In order to develop the interface tools, it is essential to determine the information that needs to be interchanged between the two systems. The primary goals of this thesis are to study the various models, determine the information that needs to be exchanged, and to develop tools to automatically extract the desired information from each model. / Master of Science

O objetivo desta tese de doutorado foi desenvolver modelos de previsão de desempenho de pavimentos asfálticos para a malha rodoviária do Estado da Bahia. Os modelos de desempenho foram desenvolvidos para pavimentos em concreto asfáltico usinado a quente e em tratamento superficial duplo. Foram utilizadas informações de um banco de dados de gerência de rodovias do DERBA, coletados em 2004, e foram coletados dados complementares de irregularidade longitudinal em 2009, após o desenvolvimento de um planejamento fatorial específico para essa finalidade. Foram utilizadas duas formas de análise estatística dos dados para avaliar a significância dos fatores e definir os parâmetros dos modelos de desempenho: a Análise Exploratória de Dados (AED) e a Análise da Variância (ANOVA). Os fatores considerados foram a idade do pavimento, o tráfego e a pluviometria e os resultados da análise exploratória de dados corroboraram os resultados encontrados pela análise da variância, tanto em relação aos efeitos dos fatores como das interações. Os modelos de previsão do desempenho foram desenvolvidos para as variáveis dependentes Índice de Condição do Pavimento Flexível (ICPF) e Área Total de Trincas (ATT), para os pavimentos em concreto asfáltico usinado a quente, e os modelos de previsão de desempenho de Área de Desgaste (ADesg) e Irregularidade Longitudinal (IRI BA), para os trechos em tratamento superficial duplo. Para a irregularidade longitudinal, também foram desenvolvidos modelos de desempenho em separado, para regiões de clima semiárido a seco (IRIBAs) e clima subúmido a úmido (IRI BAÚ). Para a verificação da adequação dos modelos de previsão de desempenho foram utilizadas a análise de resíduos, o coeficiente de correlação (r) e o coeficiente de determinação (\'R POT.2\'). Os modelos de desempenho, desenvolvidos neste trabalho para a Área Total de Trinca (ATT) e para a Irregularidade Longitudinal, foram comparados com modelos de desempenho empíricos, utilizados em programas de gerência de pavimentos por organismos rodoviários, do Brasil e do exterior, e apresentaram melhores resultados. O modelo de Irregularidade Longitudinal (IRI BA) para a rede de rodovias em tratamento superficial duplo também foi comparado com modelo desenvolvido pelo Departamento de Infraestrutura de Transportes da Bahia - DERBA, com base em levantamentos para a priorização de trechos candidatos a atividades de manutenção e reabilitação, e apresentou melhor ajuste aos dados observados. No entanto, para que sejam obtidos ainda melhores resultados, quanto à previsão de desempenho dos pavimentos da rede de rodovias pavimentadas do Estado da Bahia, é necessário que sejam realizados novos levantamentos para alimentação do banco de dados, sugerindo-se estudos complementares que incluam fatores que não puderam ser considerados neste trabalho, como a capacidade estrutural, por exemplo. Recomenda-se que o método utilizado neste trabalho, baseado no planejamento fatorial, na Análise Exploratória de Dados e na Análise de Variância, seja repetido, pois se mostrou capaz de reduzir os custos da coleta de dados sem comprometer a precisão e acurácia dos modelos de desempenho obtidos. / The objective of this thesis was to develop performance prediction models for asphalt pavements in the State of Bahia. The performance models were developed for hot-mix asphalt (HMA) and double surface treatment (ST). Information was obtained from a database maintained by DERBA (State of Bahia Department of Transportation), with data collected in 2004, and from additional collection of data, mainly of pavement roughness, performed in 2009 and that was based on an experimental design for this specific purpose. Two statistical analyses were used to assess the significant factors and define the parameters of the performance models: Exploratory Data Analysis (EDA) and Analysis of Variance (ANOVA). The factors considered were age of the pavement, traffic volume and rainfall, and the results of the exploratory data analysis corroborated the results found by the analysis of variance, both in relation to the effects of factors such as of their interactions. The performance prediction models were developed for the dependent variables Pavement Condition Index (PCI) and Area of Total Cracks (ATC) for HMA pavements, and Raveling (R) and roughness (IRI BA) for ST pavements. For roughness, performance models were also developed considering, separately, two climate conditions: semiarid to arid (IRI BAa) and sub-humid to humid (IRI BAh). Residue analysis and calculation of correlation coefficient (r) and coefficient of determination (\'R POT.2\') were performed to verify the adequacy of performance prediction models. The performance models developed in this work for Area of Total Cracks (ATC) and roughness (IRI) were compared with empirical performance models used for pavement management in Brazilian and foreign highway agencies, and they presented the best results. The performance model for roughness (IRI BA) in ST pavements was also compared to a prediction model developed by DERBA using data collected by a maintenance and rehabilitation activities prioritization program and it presented the best fit to the observed data. However, to obtain even better results regarding the performance prediction of asphalt pavements in the State of Bahia, it is necessary to conduct new field surveys to feed the database and be used in further studies that include factors that could not be considered in this work, as structural capacity, for example. It is recommended that the method used in this work, based on factorial design, Exploratory Data Analysis and ANOVA, be repeated, because it proved capable of reducing the costs of data collection without compromising precision and accuracy of the obtained performance prediction models.

Esta tese de doutorado analisa o impacto gerado pela virtualização assistida por hardware, de terceira geração, no desempenho de aplicações baseadas em SOA do tipo CPU-bound e memory-intensive, bem como nas operações de marshalling e unmarshalling de mensagens SOAP executadas nas bibliotecas dos servidores de aplicação. A partir dessa análise, concluiu-se que os recursos disponibilizados pela plataforma de hardware, principalmente no gerenciamento de memória, eliminam a sobrecarga gerada pela virtualização baseada em tradução binária, elevando os hypervisores bare-metal a um novo patamar, onde aplicações executadas sobre tais virtualizadores obtém, frequentemente, desempenho superior àquelas executadas em plataformas não virtualizadas. Assim, modificações nos modelos de desempenho de aplicações baseadas em SOA atualmente em uso são propostas, de acordo com a análise desenvolvida, para que reflitam os resultados obtidos de forma realista / This PhD thesis analyzes the impact caused by third generation hardware-assisted virtualization on the performance of CPU-bound and memory-intensive SOA-based applications, as well as the impact caused on the marshaling and unmarshalling operations of the SOAP messages executed on the libraries of the application servers. From this analysis is concluded that the resources provided by the hardware platform, mainly considering the memory management, eliminate the overload generated by virtualization based on binary translation, raising the bare-metal hypervisors to a new level, where applications running on these hypervisors reach better performance when compared to those applications running on platforms not virtualized. Thus, modifications on the performance models for SOAbased applications currently in use are proposed accordingly to the analysis developed, aiming at reflecting the results obtained in a realistic fashion