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

Unscheduled aircraft maintenance causes a lot problems and costs for aircraft operators. This is due to the fact that aircraft cause significant costs if flights have to be delayed or canceled and because spares are not always available at any place and sometimes have to be shipped across the world. Reducing the number of unscheduled maintenance is thus a great costs factor for aircraft operators. This thesis describes three methods for aircraft health monitoring and prediction; one method for system monitoring, one method for forecasting of time series and one method that combines the two other methods for one complete monitoring and prediction process. Together the three methods allow the forecasting of possible failures. The two base methods use decision trees for decision making in the processes and genetic optimization to improve the performance of the decision trees and to reduce the need for human interaction. Decision trees have the advantage that the generated code can be fast and easily processed, they can be altered by human experts without much work and they are readable by humans. The human readability and modification of the results is especially important to include special knowledge and to remove errors, which the automated code generation produced.

90% of the data that exists today has been created over the last two years only. Part of the data space is created and collected by machines, sending logs of internal measurements to be analyzed and used to evaluate service incidents. However, efficiently comprehending datasets requires more than just access to data, as Richard Hamming puts it; "The purpose of computing is insight, not numbers." A tool to simplify apprehension of complex datasets is information visualization, which works by transforming layers of information into a visual medium, enabling the human perception to quickly extract valuable information and recognise patterns. This was an experimental design-oriented research study, set out to explore if an information visualization of microservice architecture relations combined with system health data could help developers at a Swedish digital rights management company (DRMC) to find root cause incidents, increase observability and decision support, i.e. simplifying the incident handling process. To explore this, a prototype was developed and user tests consisting of a set of tasks as well as a semi-structured interview was executed by ten developers at DRMC. The results concluded that the proposed solution provided a welcomed overview of service health and dependencies but that it lacked the ability to effectively focus on certain services, essentially making it difficult to find root causes. Visualizations like this seems to be best suited for overview-, rather than focused, comprehension. Further research could be conducted on how to efficiently render large complex datasets while maintaining focus and how to account for external factors. / 90% av alla data som finns idag har skapats under de senaste två åren. En del av datautrymmet skapas och samlas in av maskiner som genererar loggar innehållandes interna systemmätningar för att utvärdera felaktiga tjänster. För att effektivt förstå ett dataset krävs mer än bara tillgång till data, som Rickard Hamming har sagt; “Syftet med datoranvändning är insikt, inte siffror.” Ett verktyg för att förenkla ens uppfattning av komplexa dataset är informationsvisualisering. Det fungerar genom att transformera lager av information till ett visuellt medium, och på så sätt tillåta mänsklig perception att snabbt extrahera värdefull information och utläsa mönster. Det här var en experimentell, design-orienterad, forskningsstudie med syftet att utforska ifall en informationsvisualisering av mikrotjänsters relationer kombinerat med system-hälso-data kunde hjälpa utvecklare på ett svenskt förvaltningsföretag av digitala rättigheter (DRMC) att hitta grundorsaken till felaktiga mikrotjänster samt utöka observerbarhet och beslutstöd, d.v.s. förenkla felhanteringsprocessen. För att utforska detta problem så utvecklades en prototyp som testades genom att låta tio utvecklare på DRMC utföra ett antal olika uppgifter samt svara på en semi-strukturerad intervju. Resultatet visade på att den föreslagna lösningen möjliggjorde en välkommen överblick över systemets hälsa och relationer, men också att den saknade möjligheten att effektivt fokusera på specifika tjänster, vilket ledde till att grundorsaksproblem var svåra att hitta. Visualiseringar som denna verkar fungera bäst för att presentera en överblick av ett system, snarare än ett fokus på specifika tjänster. Framtida forskning skulle kunna utföras för att utreda hur visualiseringar effektivt kan återge komplexa dataset utan att förlora fokus på specifika delar, samt hur externa faktorer kan integreras.

The FDR (Frequency Disturbance Recorder) is a data acquisition device for the power system. The device is portable and can be used with any residential wall outlet for frequency data collection. Furthermore, the FDR transmits calculated frequency data to the web for access by authorized users via Ethernet connection. As a result, Virginia Tech implemented Frequency Monitoring Network (FNET) with these FDR devices. FNET is a collection of identical FDRs placed in different measurement sites to allow for data integration and comparison. Frequency is an important factor for power system control and stabilization. With funding and support provided by ABB, TVA and NSF the FDRs are placed strategically all over the United States for frequency analysis, power system protection and monitoring. The purpose of this study is to refine the current FDR hardware design and establish a new design that will physically fit all the components on one Printed Circuit Board (PCB). At the same time, the software that is to be implemented on the new board is to be kept similar if not the same as that of the current design. The current FDR uses the Axiom CME555 development board and it is interfaced to the external devices through its communication ports. Even through the CME555 board is able to meet the demands of the basic FDR operations, there are still several problems associated with this design. This paper will address some of those hardware problems, as well as propose a new board design that is specifically aimed for operations of FDR. / Master of Science

La surveillance étroite d’un grand équipement physique mobile sur une trajectoire unique connue est susceptible d’avoir des applications concrètes importantes pour la sécurité des utilisateurs et de l’équipement lui-même, pour sa maintenance et pour l’optimisation du service rendu. Cette surveillance nécessite d’instrumenter l’équipement. Cette instrumentation à l’aide d’un réseau de capteurs pris sur étagère constitue l’application de l’étude menée dans cette thèse. Les problématiques de recherche concernent trois domaines : la topologie des réseaux de capteurs mobiles, le routage et la localisation. Elles sont situées par rapport à l’état de l’art. Puis une étude approfondie de l’architecture du réseau pour sur un système physique parcourant une trajectoire circulaire a été menée. La grande roue de Singapour (Singapore Flyer) est utilisée comme système type. Deux sortes de topologie ont été étudiées : une dans laquelle la liaison système mobile – partie fixe au sol est située au niveau du ou des capteurs, l’autre pour laquelle la liaison système mobile - partie fixe au sol est située au niveau du puits. Les meilleurs placements des relais ont été proposés ainsi que des algorithmes de routage géographique adaptés à chaque cas. Les performances du réseau sont mesurées par simulation. Elles montrent un bon fonctionnement pour des phénomènes dynamiques assez rapides et permettent de comparer les architectures entre-elles. Enfin une expérimentation de localisation des nœuds à base de RSSI a été menée sur un site industriel / The close supervision of a large physical equipment moving on a single known trajectory is likely to have important practical applications for the safety of users and the equipment itself, for maintenance and optimization of the provided service. This requires deploying monitoring instruments on equipment. This instrumentation using a sensor network taken off the shelf is the application of the study conducted in this thesis. The research problems address three areas: the topology of mobile sensor networks, routing and localization. They are situated in relation to the state of the art. Then a detailed study of the network architecture for a physical system moving on a circular path was conducted. Giant observation wheel of Singapore (Singapore Flyer) is used as a type system. Two kinds of topology were considered: one in which the connection mobile system - fixed part at the ground is located at sensor interface, one for which the link mobile system - fixed part at the ground is located at sink interface. The best placements of the relays have been proposed as well as geographic routing algorithms adapted to each case. The network performance is measured by simulation. The results show a correct working for fairly rapid dynamic phenomena and allow to compare architectures with each other. Finally an experiment of node localization based on RSSI was conducted on an industrial site

<p style="text-align: justify">Doktorska disertacija se bavi aktuelnom problematikom dono&scaron;enja odluka u procesu prepoznavanja tipa funkcije pouzdanosti brodskih postrojenja. U uvodnom delu doktorata obja&scaron;njeni su motivi za rad na doktorskoj disertaciji. Motivacija je proiza&scaron;la iz nastojanja da se unapredi teorija i praksa dono&scaron;enja odluka u prepoznavanju tipa funkcije pouzdanosti brodskih postrojenja, koja se zasniva na analitič kim i matematič ko-statistič ikim metodama. Pristup za re&scaron;avanje tog problema, koji je predložen u doktoratu, svodi se na teorijsku analizu radova i tehnologija, kao i na primenu sistemske dinamike i simulacione tehnologije u eksploatacionom istraž ivanju, a koje se bave problematikom klasifikacije signala sa pragom za dono&scaron;enje odluka o pouzdanosti brodskih postrojenja. To je uzrokovalo da se autor ove disertacije bavi istraživanjem novih metoda koje se mogu primeniti u oblasti pouzdanosti brodskih postrojenja. U doktoratu su razvijeni matematički i verbalni modeli pona&scaron;anja sistema pomoću signala na izlazu sistema za nekoliko slučajeva. U zaključku disertacije sistematično se interpretiraju rezultati istraživanja, potvrđuju hipoteze i komentari&scaron;u teorijska i praktična re&scaron;enja, daje se pregled mogućnosti daljih istraživanja re&scaron;enja koja su predložena disertacijom. Na primeru merenih i separiranih signala eksperimentalno, pomoću simulacione metodologije i sistemske dinamike, verifikovana je hipoteza statistič ke nezavisnosti izvornih signala. Sintezom teorijskih i analitičkih saznanja i numerički potkrepljenim praktič nim rezultatima (simulacijama), težilo se osmi&scaron;ljavanju preporuka za usklađivanje procesa dono&scaron;enja odluka na bazi prepoznavanja oblika pouzdanosti sa ukupnom sigurno&scaron;ću brodskih postrojenja, na način kako se do sada nije činilo.</p> / <p>This doctoral thesis deals with up-to-date problems related to decision making in the<br />process of recognizing forms of reliability of marine plants. Marine propulsion plant is very<br />complex and includes a set of interrelated and dependent subsystems with a large number of<br />components. The reliability and availability of such complex system depends on reliability of its<br />all components.<br />In the introductory part motives for work on this thesis are explained. Motivation has<br />come out of intention to develop the theory and practice of decision making in recognizing forms<br />of reliability of marine plants based on analytical and mathematical-statistical methods. The<br />approach to solving the problem, that is proposed in the thesis, corresponds to the theoretical<br />analysis of works and technologies as well as to the application of system dynamics and<br />simulation technology in exploitation research that again deal with the problems of classification<br />of signals with the level of decision making on the reliability of marine plants. The above<br />mentioned has lead the author to devote himself to researching new methods that can be applied<br />in the field of reliability of marine plants. The thesis presents developed mathematical and verbal<br />models of system behaviour with the use of signals observed at the system output for several<br />cases. In the conclusion of the thesis results of the research are systematically interpreted,<br />hypotheses are proved and theoretical and practical solutions commented on. Furthermore, a<br />range of possibilities for further research of solutions proposed in the thesis are presented. On the<br />example of measured and separated signals the hypothesis of statistical independence of original<br />signals has been verified experimentally using simulation methodology and system dynamics.<br />Synthesis of theoretical and analytical newly gained knowledge and numerically supported<br />practical results (simulations) have been used with the aim to reach reccomendations for<br />matching the process of decision making based on recognizing forms of reliability with total<br />safety of marine plants in a manner not used so far.</p>

The ability to implement Wide Area Monitoring and Control in power systems is developing into a need in order to prevent wide scale cascading outages. Monitoring of events in the power system provides a great deal of insight into the behaviour of the system. The research work presented in this thesis focussed on two tools that aid in monitoring: State Estimation and Synchronised Phasors provided by Phasor Measurement Units (PMU). State Estimation is essentially an on-line data processing scheme used to estimate the best possible state (i.e. voltage phasors) from a monitored set of measurements (active and reactive powers/voltage phasor measurements). The ever growing complexity and developments in the state of art calls for robust state estimators that converge accurately and rapidly. Newton’s method forms the basis for most of the solution approaches. For real-time application in modern power systems, the existing Newton-based state estimation algorithms are too fragile numerically. It is known that Newton’s algorithm may fail to converge if the initial nominal point is far from the optimal point. Sometimes Newton’s algorithm can converge to a local minima. Also Newton’s step can fail to be a descent direction if the gain matrix is nearly singular or ill-conditioned. This thesis proposes a new and more robust method that is based on linear programming and trust region techniques. The proposed formulation is suitable for Upper Bound Linear Programming. The formulation is first introduced and its convergence characteristics with the use of Upper Bound Linear Programming is studied. In the subsequent part, the solution to the same formulation is obtained using trust region algorithms. Proposed algorithms have been tested and compared with well known methods. The trust region method-based state estimator is found to be more reliable. This enhanced reliability justifies the additional time and computational effort required for its execution. One of the key elements in the synchrophasor based wide area monitoring is the Phasor Measurement Unit. Synchronized, real time, voltage phasor angle, phasor measurements over a distributed power network presents an excellent opportunity for major improvements in power system control and protection. Two of the most significant applications include state estimation and instability prediction. In recent years, there has been a significant research activity on the problem of finding the suitable number of PMUs and their optimal locations. For State Estimation, such procedures, which basically ensure observability based on network topology, are sufficient. However for instability prediction, it is very essential that the PMUs are located such that important/vulnerable buses are also directly monitored. In this thesis a method for optimal placement of PMUs, considering the vulnerable buses is developed. This method serves two purposes viz., identifying optimal locations for PMU (planning stage), and identifying the set PMUs to be closely monitored for instability prediction. The major issue is to identify the key buses when the angular and voltage stability prediction is taken into account. Integer Linear Programming technique with equality and inequality constraints is used to find out the optimal placement set. Further, various aspects of including the Phasor Measurements in state estimation algorithms are addressed. Studies are carried out on various sample test systems, an IEEE 30-bus system and real life Indian southern grid equivalents of 24-bus system, 72-bus system and 205-bus system.

The research topics presented in this dissertation focus on validation of customer-level voltage synchrophasor data for transmission system analysis, detection and categorization of power system events as measured by phasor measurement units (PMUs), and identification of the influence of power system conditions (wind power, daily and seasonal load variation) on low-frequency oscillations. Synchrophasor data can provide information across entire power systems but obtaining the data, handling the large dataset and developing tools to extract useful information from it is a challenge. To overcome the challenge of obtaining data, an independent synchrophasor network was created by taking synchrophasor measurements at customer-level voltage. The first objective is to determine if synchrophasor data taken at customer-level voltage is an accurate representation of power system behavior. The validation process was started by installing a transmission level (69 kV) PMU. The customer-level voltage measurements were validated by comparison of long term trends and low-frequency oscillations estimates. The techniques best suited for synchrophasor data analysis were identified after a detailed study and comparison. The same techniques were also applied to detect power system events resulting in the creation of novel categories for numerous events based on shared characteristics. The numerical characteristics for each category and the ranges of each numerical characteristic for each event category are identified. The final objective is to identify trends in power system behavior related to wind power and daily and seasonal variations by utilizing signal processing and statistical techniques. / text

To view the electricity supply in our society as just sockets mountedin our walls with a constant voltage output is far from the truth. Inreality, the power system supplying the electricity or the grid, is themost complex man-made dynamical system there is. It demands severecontrol and safety measures to ensure a reliable supply of electric power.Throughout the world, incidents of widespread power grid failures havebeen continuously reported. The state where electricity delivery to customersis terminated by a disturbance is called a blackout. From a stateof seemingly stable operating conditions, the grid can fast derail intoan uncontrollable state due to cascading failures. Transmission linesbecome automatically disconnected due to power flow redirections andparts of the grid become isolated and islands are formed. An islandedsub-grid incapable of maintaining safe operation conditions experiencesa blackout. A widespread blackout is a rare, but an extremely costlyand hazardous event for society.During recent years, many methods to prevent these kinds of eventshave been suggested. Controlled islanding has been a commonly suggestedstrategy to save the entire grid or parts of the grid from a blackout.Controlled islanding is a strategy of emergency control of a powergrid, in which the grid is intentionally split into a set of islanded subgridsfor avoiding an entire collapse. The key point in the strategy is todetermine appropriate separation boundaries, i.e. the set of transmissionlines separating the grid into two or more isolated parts.The power grid exhibits highly nonlinear response in the case oflarge failures. Therefore, this thesis proposes a new controlled islandingmethod for power grids based on the nonlinear Koopman Mode Analysis(KMA). The KMA is a new analyzing technique of nonlinear dynamicsbased on the so-called Koopman operator. Based on sampled data followinga disturbance, KMA is used to identify suitable partitions of thegrid.The KMA-based islanding method is numerically investigated withtwo well-known test systems proposed by the Institute of Electrical andElectronics Engineers (IEEE). By simulations of controlled islanding inthe test system, it is demonstrated that the grid’s response following afault can be improved with the proposed method.The proposed method is compared to a method of partitioning powergrids based on spectral graph theory which captures the structural propertiesof a network. It is shown that the intrinsic structural propertiesof a grid characterized by spectral graph theory are also captured by theKMA. This is shown both by numerical simulations and a theoreticalanalysis.

Auditory preemption theory suggests two competing assumptions for the attention-capturing and performance-altering properties of auditory tasks. In onset preemption, attention is immediately diverted to the auditory channel. Strategic preemption involves a decision process in which the operator maintains focus on more complex auditory messages. The limitation in this process is that the human auditory, or echoic, memory store has a limit of 2 to 5 seconds, after which the message must be processed or it decays. In contrast, multiple resource theory suggests that visual and auditory tasks may be efficiently time-shared because two different pools of cognitive resources are used. Previous research regarding these competing assumptions has been limited and equivocal. Thus, the current research focused on systematically examining the effects of complexity and timing of communication interruptions on visual detection tasks. It was hypothesized that both timing and complexity levels would impact detection performance in a multi-task environment. Study 1 evaluated the impact of complexity and timing of communications occurring before malfunctions in an ongoing visual detection task. Twenty-four participants were required to complete each of the eight timing blocks that included simple or complex communications occurring simultaneously, and at 2, 5, or 8 seconds before detection events. For simple communications, participants repeated three pre-recorded words. However, for complex communications, they generated three words beginning with the same last letter of a word prompt. Results indicated that complex communications at two seconds or less occurring before a visual detection event significantly impacted response time with a 1.3 to 1.6 second delay compared to all the other timings. Detection accuracy for complex communication tasks under the simultaneous condition was significantly degraded compared to simple communications at five seconds or more prior to the task. This resulted in a 20% decline in detection accuracy. Additionally, participants' workload ratings for complex communications were significantly higher than simple communications. Study 2 examined the timing of communications occurring at the corresponding seconds after the visual detection event. Twenty-four participants were randomly assigned to the communication complexity and timing blocks as in study 1. The results did not find significant performance effects of timing or complexity of communications on detection performance. However the workload ratings for the 2 and 5 second complex communication presentations were higher compared to the same simple communication conditions. Overall, these findings support the strategic preemption assumption for well-defined, complex communications. The onset preemption assumption for simple communications was not supported. These results also suggest that the boundaries of the multiple resource theory assumption may exist up to the limits of the echoic memory store. Figures of merit for task performance under the varying levels of timing and complexity are presented. Several theoretical and practical implications are discussed.