Submitted by Fabio Sobreira Campos da Costa (fabio.sobreira@ufpe.br) on 2016-04-05T15:06:42Z No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Sidartha Carvalho Dissertacao de Mestrado.pdf: 2847376 bytes, checksum: e22fe197fc67ebb8e11ceb85baa66bc8 (MD5) / Made available in DSpace on 2016-04-05T15:06:42Z (GMT). No. of bitstreams: 2 license_rdf: 1232 bytes, checksum: 66e71c371cc565284e70f40736c94386 (MD5) Sidartha Carvalho Dissertacao de Mestrado.pdf: 2847376 bytes, checksum: e22fe197fc67ebb8e11ceb85baa66bc8 (MD5) Previous issue date: 2015-02-26 / FACEPE / Diante do crescimento das vendas de smartphones, em específico com o sistema operacional Android, em suas variadas versões e diversidade de aparelhos, um problema chamado de fragmentação foi destacado. Com o incremento de dispositivos multinúcleos e multiprocessadores, crescem as restrições na potência e ocorrência de superaquecimento. Alguns trabalhos consultados se especializam em resolver o problema da fragmentação, porém não se preocupam com dados energéticos, enquanto outros somente relatam técnicas que permitem a diminuição no consumo energético, mas nenhum possibilita a integração de testes multidispositivos, com o intuito de solucionar o problema da fragmentação, com foco energético. Este trabalho oferece a modelagem e implementação de um ambiente de medição que auxilia na avaliação do consumo energético de dispositivos com o sistema operacional Android. O ambiente proposto permite que seja feita uma análise energética de dispositivos em diferentes frequências no processador e redes de dados de forma simultânea e com manipulação pela Web. Uma infraestrutura genérica de medição, a qual permite a coleta de tensão e corrente de dispositivos alimentados por bateria foi criada juntamente com um canal que permite a comunicação com um computador para análise dos dados. No tangente ao software, foi construída uma plataforma Web para manipulação de componentes do dispositivo e replicação de testes de forma automatizada. Para testar o uso do ambiente proposto, foram executados oito testes envolvendo Navegação Web, Streaming de Vídeo e Carga de Processamento nos smartphones disponíveis. Os resultados dos testes são apresentados e discutidos em detalhe. / Recently we can observe a growth in smartphone sales, in particular those running Android operating system. In several versions and devices, a problem known as fragmentation was featured. Along with the increase of multi-core and multi-processor devices, growing restrictions on power and overheating were highlighted. Some studies focus on solving the fragmentation problem and do not care about energy analysis, while other reports only techniques which allow a reduction in energy consumption, but none enables the integration of multi-devices tests in order to solve the problem of fragmentation with energy focus. This study provides the modeling and implementation of a measurement environment that helps the evaluation of the energy consumption of devices with the Android operating system. The proposed environment makes possible an analysis of the energy in devices at different processor frequencies and data networks on multiple devices simultaneously with Web access. A generic and cheap infrastructure measurement, which allows the collection of voltage and current of battery-powered devices and a channel that allows communication with a computer for data analysis were built. In regard to the software, a Web platform for manipulation of device peripherals and automated tests replication was constructed. In order to test the use of the proposed environment, eight tests were performed covering Web Browsing, Video Streaming and CPU load on the available smartphones, and then the results were explained and discussed.

Ce mémoire présente une étude expérimentale de la fragmentation en phase gazeuse des biomolécules, adénine (H5C5N5) et porphyrine FeTPPCl (C44H28N4FeCl), induite par collision avec des ions à basse énergie. La distribution de population pour chaque voie de dissociation a été mesurée en fonction de l'énergie d'excitation des ions moléculaires parents avec la méthode CIDEC (Collision Induced Dissociation under Energy Control). Dans les collisions entre Cl+ à 3keV et adénine (Ade), le schéma de fragmentation de Ade2+ est dominée par la perte de H2CN+ et les émissions successives de HCN. La distribution de l'énergie des Ade2+ parents confirme la dynamique des émissions successives. Une voie de dissociation spécifique, à savoir l'émission successive de H2CN+ et HC2N2 est observée. Les schémas de fragmentation des ions moléculaires FeTPPCl1+, 2+, 3+ sont étudiés dans des collisions avec Kr8+ à 80 keV. Il est constaté qu’indépendante de l'état de charge initiale de FeTPPClr+ (r=1, 2, 3), la perte de Cl0 constitue la première étape de la chaîne de dissociation, tandis que l’état de charge initiale des molécules joue un rôle important dans les étapes suivantes de la dissociation. Dans les collisions avec H+ et F+ à 3keV, dû à un effet de fenêtre de réaction dans les processus de production d’ions négatifs, des schémas de fragmentation très différents sont observés pour FeTPPCl2+. Grâce à la mesure de l’énergie interne des molécules parents, la perte de nH2 est observée et analysée. De plus, le rendement de production d'ions négatifs, mesuré à environ 1% dans des collisions F2+-Ade à 30 keV, est étudié dans ce travail en utilisant une nouvelle approche expérimentale. / In this work, the Collision Induced Dissociation under Energy Control method was extended to study the fragmentation of gas-phase biomolecules adenine (H5C5N5) and porphyrin FeTPPCl (C44H28N4FeCl). The population distribution for each dissociation channel has been experimentally determined as a function of the excitation energy of the parent molecular ions at a well-determined initial charge state. In collisions between Cl+ and adenine (Ade) at 3keV, the fragmentation pattern of Ade2+ is dominated by the loss of H2CN+ and the successive emission of HCN. The energy distribution of the parent dications confirms the successive emission dynamics. A specific decay channel is observed, i.e., the emission of a charged H2CN+ followed by the emission of HC2N2. In Kr8+-FeTPPCl collisions at 80keV, parent ions FeTPPCl1+,2+,3+ are observed, along with the corresponding decay patterns. It is found that in the first step the dominant low-energy-cost decay channel is the emission of Cl0 independent of the initial charge state of FeTPPClr+ (r=1-3). For the resulted dication FeTPP2+, the dominant fragmentation channel is the neutral evaporation; for the trication however, the dominant fragmentation channel is the asymmetrical fission. In the case of H+ and F+ impact at 3keV, due to the different reaction windows opened in the two collision systems, different fragmentation patterns are observed. Furthermore, nH2 loss processes are observed. Additionally, the production yield of the negative ion emerged in F2+-Ade collision at 30keV is measured to be about 1% using a new experimental approach.

In this study eddy covariance was used to measure sensible heat, latent heat, and carbon dioxide fluxes for the months of August, September, and October of 2009 within the roughness sublayer (RSL) of the urban center of Portland, OR. Vehicle traffic and solar radiation were also measured for the month of October. Flux measurements were compared with measurements from other urban areas as a test of reasonableness. CO₂ fluxes were nearly always positive and were strongly correlated with the weekday diurnal traffic cycle. CO₂ fluxes averaged 6.6 μmol/m^²s, which is less than other published measurements in urban areas. Sensible and latent heat fluxes followed the expected diurnal profile associated with solar radiation. Average sensible heat flux decreased as the season changed from summer to fall, moving from an average of 39 W/m^² in August to 12 W/m^² in October. A corresponding increase in latent heat flux was observed during this period, changing from an average of 10 W/m^² in August to 17 W/m^² in October. Heat flux behavior and amplitude was consistent with other urban measurements, though amplitude varies considerably from city to city. Stationarity was shown to positively influence measured CO₂ fluxes, but to have little effect on measured heat fluxes. Preliminary comparisons of October sensible heat and CO₂ fluxes to an inventory-based estimate of vehicle emissions indicate that eddy covariance measurements underestimate the true fluxes by 50%.

This thesis applies energy measurements to the domain of continuous integration (CI) and continuous delivery (CD) pipelines. The goal is to conduct transparent and fine-granular energy measurements of these pipelines, increasing awareness and allowing optimizations regarding their energy efficiency. CI and CD automate processes like compilation, running tests, and code analysis tools and can improve the software quality and developer experience and enable more frequent releases. Initially, the applicability of existing energy measurement approaches for these tasks is analyzed. Afterward, a generic framework consisting of a pipeline run analyzer, a resource consumption collector, and an energy calculator is proposed. A representative implementation for a state-of-the-art infrastructure is devised to demonstrate its functionality, enabling the collection, analysis, and interpretation of data from real-world examples. Finally, it is examined whether this data aligns with the theoretical considerations and can be used to optimize the pipelines. The overall goal is to contribute to the sustainability of DevOps processes and therefore counteract the disastrous consequences of unrestrained climate change. / Denna avhandling tillämpar energimätningar på området kontinuerlig integration (CI) och kontinuerlig leverans (CD). Målet är att genomföra transparenta och finkorniga energimätningar av dessa pipelines, vilket ökar medvetenheten och möjliggör optimeringar av deras energieffektivitet. CI och CD automatiserar processer som kompilering, testkörning och kodanalysverktyg och kan förbättra programvarukvaliteten och utvecklarens upplevelse samt möjliggöra tätare lanseringar. Inledningsvis analyseras tillämpligheten av befintliga metoder för energimätning för dessa uppgifter. Därefter föreslås ett generiskt ramverk som består av en analysator för pipelinekörning, en insamlare av resursförbrukning och en energikalkylator. För att demonstrera dess funktionalitet utarbetas en representativ implementering för en modern infrastruktur som möjliggör insamling, analys och tolkning av data från verkliga exempel. Slutligen undersöks om dessa uppgifter stämmer överens med de teoretiska övervägandena och kan användas för att optimera rörledningarna. Det övergripande målet är att bidra till hållbarheten i DevOpsprocesser och därmed motverka de katastrofala konsekvenserna av ohämmade klimatförändringar.

<p>U disertaciji je prikazan i predložen novi metod merenja snage<br />i energije vetra koji se bazira na stohastičkoj digitalnoj mernoj<br />metodi (SDMM), umesto do sada korišćene standardne sempling<br />metode (SSM). Standard za merenje brzine vetra predstavlja<br />anemometar sa šoljicama, koji na svom izlazu daje sinusni napon<br />čija su amplituda i frekvencija linearno zavisne od brzine<br />vetra. Kako su snaga i energija vetra proporcionalne trećem<br />stepenu brzine vetra, u tu svrhu se primenjuje generalizovan<br />hardver namenjen za računanje proizvoda tri ulazna signala. Zbog<br />niskog frekvencijskog opsega ulaznog signala izvršeno je<br />značajno pojednostavljenje hardvera mernog uređaja. Teorijskom<br />analizom, simulacijom, a potom i eksperimentom potvrđena je<br />hipoteza istraživanja, da se SDMM uspešno može koristiti za<br />merenje snage i energije vetra.</p> / <p>In this dissertation а new wind power and energy measurement<br />method, based on Stochastic Digital Measurement Method (SDMM),<br />was elaborated and proposed, to replace the commonly used Standard<br />Sampling Method (SSM). The standard for wind speed measurement is<br />a cup anemometer that provides a sinusoidal voltage on its output,<br />having the amplitude and frequency linearly dependent of the wind<br />speed. Since the wind power and energy are proportional to the third<br />degree of the wind speed, the general case hardware has been designed<br />for the product calculation over three input signals. The low frequency<br />range of the input signal allows for a significant simplification of the<br />measuring instrument hardware. By theoretical analysis and<br />simulation, and consequently through an experiment the hypothesis of<br />this research was confirmed - that SDMM can be successfully used to<br />measure wind power and energy.</p>

Energy efficiency is a major criterion for computing in general and High Performance Computing in particular. When optimizing for energy efficiency, it is essential to measure the underlying metric: energy consumption. To fully leverage energy measurements, their quality needs to be well-understood. To that end, this thesis provides a rigorous evaluation of various energy measurement techniques. I demonstrate how the deliberate selection of instrumentation points, sensors, and analog processing schemes can enhance the temporal and spatial resolution while preserving a well-known accuracy. Further, I evaluate a scalable energy measurement solution for production HPC systems and address its shortcomings. Such high-resolution and large-scale measurements present challenges regarding the management of large volumes of generated metric data. I address these challenges with a scalable infrastructure for collecting, storing, and analyzing metric data. With this infrastructure, I also introduce a novel persistent storage scheme for metric time series data, which allows efficient queries for aggregate timelines. To ensure that it satisfies the demanding requirements for scalable power measurements, I conduct an extensive performance evaluation and describe a productive deployment of the infrastructure. Finally, I describe different approaches and practical examples of analyses based on energy measurement data. In particular, I focus on the combination of energy measurements and application performance traces. However, interweaving fine-grained power recordings and application events requires accurately synchronized timestamps on both sides. To overcome this obstacle, I develop a resilient and automated technique for time synchronization, which utilizes crosscorrelation of a specifically influenced power measurement signal. Ultimately, this careful combination of sophisticated energy measurements and application performance traces yields a detailed insight into application and system energy efficiency at full-scale HPC systems and down to millisecond-range regions.:1 Introduction 2 Background and Related Work 2.1 Basic Concepts of Energy Measurements 2.1.1 Basics of Metrology 2.1.2 Measuring Voltage, Current, and Power 2.1.3 Measurement Signal Conditioning and Analog-to-Digital Conversion 2.2 Power Measurements for Computing Systems 2.2.1 Measuring Compute Nodes using External Power Meters 2.2.2 Custom Solutions for Measuring Compute Node Power 2.2.3 Measurement Solutions of System Integrators 2.2.4 CPU Energy Counters 2.2.5 Using Models to Determine Energy Consumption 2.3 Processing of Power Measurement Data 2.3.1 Time Series Databases 2.3.2 Data Center Monitoring Systems 2.4 Influences on the Energy Consumption of Computing Systems 2.4.1 Processor Power Consumption Breakdown 2.4.2 Energy-Efficient Hardware Configuration 2.5 HPC Performance and Energy Analysis 2.5.1 Performance Analysis Techniques 2.5.2 HPC Performance Analysis Tools 2.5.3 Combining Application and Power Measurements 2.6 Conclusion 3 Evaluating and Improving Energy Measurements 3.1 Description of the Systems Under Test 3.2 Instrumentation Points and Measurement Sensors 3.2.1 Analog Measurement at Voltage Regulators 3.2.2 Instrumentation with Hall Effect Transducers 3.2.3 Modular Instrumentation of DC Consumers 3.2.4 Optimal Wiring for Shunt-Based Measurements 3.2.5 Node-Level Instrumentation for HPC Systems 3.3 Analog Signal Conditioning and Analog-to-Digital Conversion 3.3.1 Signal Amplification 3.3.2 Analog Filtering and Analog-To-Digital Conversion 3.3.3 Integrated Solutions for High-Resolution Measurement 3.4 Accuracy Evaluation and Calibration 3.4.1 Synthetic Workloads for Evaluating Power Measurements 3.4.2 Improving and Evaluating the Accuracy of a Single-Node Measuring System 3.4.3 Absolute Accuracy Evaluation of a Many-Node Measuring System 3.5 Evaluating Temporal Granularity and Energy Correctness 3.5.1 Measurement Signal Bandwidth at Different Instrumentation Points 3.5.2 Retaining Energy Correctness During Digital Processing 3.6 Evaluating CPU Energy Counters 3.6.1 Energy Readouts with RAPL 3.6.2 Methodology 3.6.3 RAPL on Intel Sandy Bridge-EP 3.6.4 RAPL on Intel Haswell-EP and Skylake-SP 3.7 Conclusion 4 A Scalable Infrastructure for Processing Power Measurement Data 4.1 Requirements for Power Measurement Data Processing 4.2 Concepts and Implementation of Measurement Data Management 4.2.1 Message-Based Communication between Agents 4.2.2 Protocols 4.2.3 Application Programming Interfaces 4.2.4 Efficient Metric Time Series Storage and Retrieval 4.2.5 Hierarchical Timeline Aggregation 4.3 Performance Evaluation 4.3.1 Benchmark Hardware Specifications 4.3.2 Throughput in Symmetric Configuration with Replication 4.3.3 Throughput with Many Data Sources and Single Consumers 4.3.4 Temporary Storage in Message Queues 4.3.5 Persistent Metric Time Series Request Performance 4.3.6 Performance Comparison with Contemporary Time Series Storage Solutions 4.3.7 Practical Usage of MetricQ 4.4 Conclusion 5 Energy Efficiency Analysis 5.1 General Energy Efficiency Analysis Scenarios 5.1.1 Live Visualization of Power Measurements 5.1.2 Visualization of Long-Term Measurements 5.1.3 Integration in Application Performance Traces 5.1.4 Graphical Analysis of Application Power Traces 5.2 Correlating Power Measurements with Application Events 5.2.1 Challenges for Time Synchronization of Power Measurements 5.2.2 Reliable Automatic Time Synchronization with Correlation Sequences 5.2.3 Creating a Correlation Signal on a Power Measurement Channel 5.2.4 Processing the Correlation Signal and Measured Power Values 5.2.5 Common Oversampling of the Correlation Signals at Different Rates 5.2.6 Evaluation of Correlation and Time Synchronization 5.3 Use Cases for Application Power Traces 5.3.1 Analyzing Complex Power Anomalies 5.3.2 Quantifying C-State Transitions 5.3.3 Measuring the Dynamic Power Consumption of HPC Applications 5.4 Conclusion 6 Summary and Outlook

The interest in using State-Of-The-Art (SOTA) Pre-Trained Language Model (PLM) in product development is growing. The fact that developers can use PLM has changed the way to build reliable models, and it is the go-to method for many companies and organizations. Selecting the Natural Language Processing (NLP) model with the highest accuracy is the usual way of deciding which PLM to use. However, with growing concerns about negative climate changes, we need new ways of making decisions that consider the impact on our future needs. The best solution with the highest accuracy might not be the best choice when other parameters matter, such as sustainable development. This thesis investigates how to calculate an approximate total cost considering Operating Expenditure (OPEX) and CO2~emissions for a deployed NLP solution over a given period, specifically the inference phase. We try to predict the total cost with Floating Point Operation (FLOP) and test NLP models on a classification task. We further present the tools to make energy measurements and examine the metric FLOP to predict costs. Using a bottom-up approach, we investigate the components that affect the cost and measure the energy consumption for different deployed models. By constructing this cost model and testing it against real-life examples, essential information about a given NLP implementation and the relationship between monetary and environmental costs will be derived. The literature studies reveal that the derival of a cost model is a complex area, and the results confirm that it is not a straightforward procedure to approximate energy costs. Even if a cost model was not feasible to derive with the resources given, this thesis covers the area and shows why it is complex by examine FLOP. / Intresset att använda State-Of-The-Art (SOTA) Pre-Trained Language Model (PLM) i produktutveckling växer. Det faktum att utvecklare kan använda PLM har förändrat sättet att träna tillförlitliga modeller på och det är den bästa metoden för många företag och organisationer att använda SOTA Naturlig Språkbehandling (NLP). Att välja NLP-modellen med högsta noggrannhet är det vanliga sättet att bestämma vilken PLM som ska användas. Men med växande oro för miljöförändringar behöver vi nya sätt att fatta beslut som kommer att påverka våra framtida behov. Denna avhandling undersöker hur man beräknar en ungefärlig totalkostnad med hänsyn till Operating Expenditure (OPEX) och CO2~utsläpp för en utplacerad NLP-lösning under en given period, dvs slutledningsfasen. Vi försöker förutspå den totala kostnaden med flyttalsoperationer och testar mot en klassificerings uppgift. Vi undersöker verktygen för att göra mätningar samt variabeln Flyttalsoperationer för att förutspå energiförbrukning.