Data mining is an increasingly important technology for extracting useful knowledge hidden in large collections of data. There are, however, negative social perceptions about data mining, among which potential privacy violation and potential discrimination. The former is an unintentional or deliberate disclosure of a user pro le or activity data as part of the output of a data mining algorithm or as a result of data sharing. For this reason, privacy preserving data mining has been introduced to trade o the utility of the resulting data/models for protecting individual privacy. The latter consists of treating people unfairly on the basis of their belonging to a speci c group. Automated data collection and data mining techniques such as classi cation have paved the way to making automated decisions, like loan granting/denial, insurance premium computation, etc. If the training datasets are biased in what regards discriminatory attributes like gender, race, religion, etc., discriminatory decisions may ensue. For this reason, anti-discrimination techniques including discrimination discovery and prevention have been introduced in data mining. Discrimination can be either direct or indirect. Direct discrimination occurs when decisions are made based on discriminatory attributes. Indirect discrimination occurs when decisions are made based on non-discriminatory attributes which are strongly correlated with biased discriminatory ones. In the rst part of this thesis, we tackle discrimination prevention in data mining and propose new techniques applicable for direct or indirect discrimination prevention individually or both at the same time. We discuss how to clean training datasets and outsourced datasets in such a way that direct and/or indirect discriminatory decision rules are converted to legitimate (non-discriminatory) classi cation rules. The experimental evaluations demonstrate that the proposed techniques are e ective at removing direct and/or indirect discrimination biases in the original dataset while preserving data quality. In the second part of this thesis, by presenting samples of privacy violation and potential discrimination in data mining, we argue that privacy and discrimination risks should be tackled together. We explore the relationship between privacy preserving data mining and discrimination prevention in data mining to design holistic approaches capable of addressing both threats simultaneously during the knowledge discovery process. As part of this e ort, we have investigated for the rst time the problem of discrimination and privacy aware frequent pattern discovery, i.e. the sanitization of the collection of patterns mined from a transaction database in such a way that neither privacy-violating nor discriminatory inferences can be inferred on the released patterns. Moreover, we investigate the problem of discrimination and privacy aware data publishing, i.e. transforming the data, instead of patterns, in order to simultaneously ful ll privacy preservation and discrimination prevention. In the above cases, it turns out that the impact of our transformation on the quality of data or patterns is the same or only slightly higher than the impact of achieving just privacy preservation.

Executable models play a key role in many development methods (such as MDD and MDA) by facilitating the immediate simulation/implementation of the software system under development. This is possible because executable models include a fine-grained specification of the system behaviour using an action language. Executable models are not a new concept but are now experiencing a comeback. As a relevant example, the OMG has recently published the first version of the “Foundational Subset for Executable UML Models” (fUML) standard, an executable subset of the UML that can be used to define, in an operational style, the structural and behavioural semantics of systems. The OMG has also published a beta version of the “Action Language for fUML” (Alf) standard, a concrete syntax conforming to the fUML abstract syntax, that provides the constructs and textual notation to specify the fine-grained behaviour of systems. The OMG support to executable models is substantially raising the interest of software companies for this topic. Given the increasing importance of executable models and the impact of their correctness on the final quality of software systems derived from them, the existence of methods to verify the correctness of executable models is becoming crucial. Otherwise, the quality of the executable models (and in turn the quality of the final system generated from them) will be compromised. Despite the number of research works targetting the verification of software models, their computational cost and poor feedback makes them difficult to integrate in current software development processes. Therefore, there is the need for efficient and useful methods to check the correctness of executable models and tools integrated to the modelling tools used by designers. In this thesis we propose a verification framework to help the designers to improve the quality of their executable models. Our framework is composed of a set of lightweight static methods, i.e. methods that do not require to execute the model in order to check the desired property. These methods are able to check several properties over the behavioural part of an executable model (for instance, over the set of operations that compose a behavioural executable model) such as syntactic correctness (i.e. all the operations in the behavioural model conform to the syntax of the language in which it is described), non-redundancy (i.e. there is no another operation with exactly the same behaviour), executability (i.e. after the execution of an operation, the reached system state is -in case of strong executability- or may be -in case of weak executability- consistent with the structural model and its integrity constraints) and completeness (i.e. all possible changes on the system state can be performed through the execution of the operations defined in the executable model). For incorrect models, the methods that compose our verification framework return a meaningful feedback that helps repairing the detected inconsistencies.

In a period of time in which the content available through the Internet increases exponentially and is more easily accessible every day, techniques for aiding the selection and extraction of important and personalised information are of vital importance. Recommender Systems (RS) appear as a tool to help the user in a decision making process by evaluating a set of objects or alternatives and aiding the user at choosing which one/s of them suits better his/her interests or preferences. Those preferences need to be accurate enough to produce adequate recommendations and should be updated if the user changes his/her likes or if they are incorrect or incomplete. In this work an adequate model for managing user preferences in a multi-attribute (numerical and categorical) environment is presented to aid at providing recommendations in those kinds of contexts. The evaluation process of the recommender system designed is supported by a new aggregation operator (Unbalanced LOWA) that enables the combination of the information that defines an alternative into a single value, which then is used to rank the whole set of alternatives. After the recommendation has been made, learning processes have been designed to evaluate the user interaction with the system to find out, in a dynamic and unsupervised way, if the user profile in which the recommendation process relies on needs to be updated with new preferences. The work detailed in this document also includes extensive evaluation and testing of all the elements that take part in the recommendation and learning processes.

Power constraints led to the end of exponential growth in single–processor performance, which characterized the semiconductor industry for many years. Single–chip multiprocessors allowed the performance growth to continue so far. Yet, Amdahl’s law asserts that the overall performance of future single–chip multiprocessors will depend crucially on single–processor performance. In a multiprocessor a small growth in single–processor performance can justify the use of significant resources. Partitioning the layout of critical components can improve the energy–efficiency and ultimately the performance of a single processor. In a clustered microarchitecture parts of these components form clusters. Instructions are processed locally in the clusters and benefit from the smaller size and complexity of the clusters components. Because the clusters together process a single instruction stream communications between clusters are necessary and introduce an additional cost. This thesis proposes the design of a distributed memory unit and first level cache in the context of a clustered microarchitecture. While the partitioning of other parts of the microarchitecture has been well studied the distribution of the memory unit and the cache has received comparatively little attention. The first proposal consists of a set of cache bank predictors. Eight different predictor designs are compared based on cost and accuracy. The second proposal is the distributed memory unit. The load and store queues are split into smaller queues for distributed disambiguation. The mapping of memory instructions to cache banks is delayed until addresses have been calculated. We show how disambiguation can be implemented efficiently with unordered queues. A bank predictor is used to map instructions that consume memory data near the data origin. We show that this organization significantly reduces both energy usage and latency. The third proposal introduces Dispatch Throttling and Pre-Access Queues. These mechanisms avoid load/store queue overflows that are a result of the late allocation of entries. The fourth proposal introduces Memory Issue Queues, which add functionality to select instructions for execution and re-execution to the memory unit. The fifth proposal introduces Conservative Deadlock Aware Entry Allocation. This mechanism is a deadlock safe issue policy for the Memory Issue Queues. Deadlocks can result from certain queue allocations because entries are allocated out-of-order instead of in-order like in traditional architectures. The sixth proposal is the Early Release of Load Queue Entries. Architectures with weak memory ordering such as Alpha, PowerPC or ARMv7 can take advantage of this mechanism to release load queue entries before the commit stage. Together, these proposals allow significantly smaller and more energy efficient load queues without the need of energy hungry recovery mechanisms and without performance penalties. Finally, we present a detailed study that compares the proposed distributed memory unit to a centralized memory unit and confirms its advantages of reduced energy usage and of improved performance.

High Performance Computing and Supercomputing is the high end area of the computing science that studies and develops the most powerful computers available. Current supercomputers are extremely complex so are the applications that run on them. To take advantage of the huge amount of computing power available it is strictly necessary to maximize the knowledge we have about how these applications behave and perform. This is the mission of the (parallel) performance analysis. In general, performance analysis toolkits oUer a very simplistic manipulations of the performance data. First order statistics such as average or standard deviation are used to summarize the values of a given performance metric, hiding in some cases interesting facts available on the raw performance data. For this reason, we require the Performance Analytics, i.e. the application of Data Analytics techniques in the performance analysis area. This thesis contributes with two new techniques to the Performance Analytics Veld. First contribution is the application of the cluster analysis to detect the parallel application computation structure. Cluster analysis is the unsupervised classiVcation of patterns (observations, data items or feature vectors) into groups (clusters). In this thesis we use the cluster analysis to group the CPU burst of a parallel application, the regions on each process in-between communication calls or calls to the parallel runtime. The resulting clusters obtained are the diUerent computational trends or phases that appear in the application. These clusters are useful to understand the behaviour of computation part of the application and focus the analyses to those that present performance issues. We demonstrate that our approach requires diUerent clustering algorithms previously used in the area. Second contribution of the thesis is the application of multiple sequence alignment algorithms to evaluate the computation structure detected. Multiple sequence alignment (MSA) is technique commonly used in bioinformatics to determine the similarities across two or more biological sequences: DNA or roteins. The Cluster Sequence Score we introduce applies a Multiple Sequence Alignment (MSA) algorithm to evaluate the SPMDiness of an application, i.e. how well its computation structure represents the Single Program Multiple Data (SPMD) paradigm structure. We also use this score in the Aggregative Cluster Re-Vnement, a new clustering algorithm we designed, able to detect the SPMD phases of an application at Vne-grain, surpassing the cluster algorithms we used initially. We demonstrate the usefulness of these techniques with three practical uses. The Vrst one is an extrapolation methodology able to maximize the performance metrics that characterize the application phases detected using a single application execution. The second one is the use of the computation structure detected to speedup in a multi-level simulation infrastructure. Finally, we analyse four production-class applications using the computation characterization to study the impact of possible application improvements and portings of the applications to diUerent hardware conVgurations. In summary, this thesis proposes the use of cluster analysis and sequence analysis to automatically detect and characterize the diUerent computation trends of a parallel application. These techniques provide the developer / analyst an useful insight of the application performance and ease the understanding of the application’s behaviour. The contributions of the thesis are not reduced to proposals and publications of the techniques themselves, but also practical uses to demonstrate their usefulness in the analysis task. In addition, the research carried out during these years has provided a production tool for analysing applications’ structure, part of BSC Tools suite.

No description available.

Nowadays, it would be very difficult to deny the need to prioritize sustainable development through energy efficiency at all consumption levels. In this context, an energy management system (EMS) is a suitable option for continuously improving energy efficiency, particularly on the user side. An EMS is a set of technological tools that manages energy consumption information and allows its analysis. EMS, in combination with information technologies, has given rise to intelligent EMS (iEMS), which, aside from lending support to monitoring and reporting functions as an EMS does, it has the ability to model, forecast, control and diagnose energy consumption in a predictive way. The main objective of an iEMS is to continuously improve energy efficiency (on-line) as automatically as possible. The core of an iEMS is its load modeling forecasting system (LMFS). It takes advantage of historical information on energy consumption and energy-related variables in order to model and forecast load profiles and, if available, generator profiles. These models and forecasts are the main information used for iEMS applications for control and diagnosis. That is why in this thesis we have focused on the study, analysis and development of LMFS on the user side. The fact that the LMFS is applied on the user side to support an iEMS means that specific characteristics are required that in other areas of load forecasting they are not. First of all, the user-side load profiles (LPs) have a higher random behavior than others, as for example, in power system distribution or generation. This makes the modeling and forecasting process more difficult. Second, on the user side --for example an industrial user-- there is a high number and variety of places that can be monitored, modeled and forecasted, as well as their precedence or nature. Thus, on the one hand, an LMFS requires a high degree of autonomy to automatically or autonomously generate the demanded models. And on the other hand, it needs a high level of adaptability in order to be able to model and forecast different types of loads and different types of energies. Therefore, the addressed LMFS are those that do not look only for accuracy, but also adaptability and autonomy. Seeking to achieve these objectives, in this thesis work we have proposed three novel LMFS schemes based on hybrid algorithms from computational intelligence, signal processing and statistical theory. The first of them looked to improve adaptability, keeping in mind the importance of accuracy and autonomy. It was called an evolutionary training algorithm (ETA) and is based on adaptivenetwork-based-fuzzy-inference system (ANFIS) that is trained by a multi-objective genetic algorithm instead of its traditional training algorithm. As a result of this hybrid, the generalization capacity was improved (avoiding overfitting) and an easily adaptable training algorithm for new adaptive networks based on traditional ANFIS was obtained. The second scheme deals with LMF autonomy in order to build models from multiple loads automatically. Similar to the previous proposal, an ANFIS and a MOGA were used. In this case, the MOGA was used to find a near-optimal configuration for the ANFIS instead of training it. The LMFS relies on this configuration to work properly, as well as to maintain accuracy and generalization capabilities. Real data from an industrial scenario were used to test the proposed scheme and the multi-site modeling and self-configuration results were satisfactory. Furthermore, other algorithms were satisfactorily designed and tested for processing raw data in outlier detection and gap padding. The last of the proposed approaches sought to improve accuracy while keeping autonomy and adaptability. It took advantage of dominant patterns (DPs) that have lower time resolution than the target LP, so they are easier to model and forecast. The Hilbert-Huang transform and Hilbert-spectral analysis were used for detecting and selecting the DPs. Those selected were used in a proposed scheme of partial models (PM) based on parallel ANFIS or artificial neural networks (ANN) to extract the information and give it to the main PM. Therefore, LMFS accuracy improved and the user-side LP noising problem was reduced. Additionally, in order to compensate for the added complexity, versions of self-configured sub-LMFS for each PM were used. This point was fundamental since, the better the configuration, the better the accuracy of the model; and subsequently the information provided to the main partial model was that much better. Finally, and to close this thesis, an outlook of trends regarding iEMS and an outline of several hybrid algorithms that are pending study and testing are presented. / En el contexto energético actual y particularmente en el lado del usuario, el concepto de sistema de gestión energética (EMS) se presenta como una alternativa apropiada para mejorar continuamente la eficiencia energética. Los EMSs en combinación con las tecnologías informáticas dan origen al concepto de iEMS, que además de soportar las funciones de los EMS, tienen la capacidad de modelar, pronosticar, controlar y supervisar los consumos energéticos. Su principal objetivo es el de realizar una mejora continua, lo más autónoma posible y predictiva de la eficiencia energética. Este tipo de sistemas tienen como núcleo fundamental el sistema de modelado y pronóstico de consumos (Load Modeling and Forecasting System, LMFS). El LMFS está habilitado para pronosticar el comportamiento futuro de cargas y, si es necesario, de generadores. Es sobre estos pronósticos sobre los cuales el iEMS puede realizar sus tareas automáticas y predictivas de optimización y supervisión. Los LMFS en el lado del usuario son el foco de esta tesis. Un LMFS en el lado del usuario, diseñado para soportar un iEMS requiere o demanda ciertas características que en otros contextos no serían tan necesarias. En primera estancia, los perfiles de los usuarios tienen un alto grado de aleatoriedad que los hace más difíciles de pronosticar. Segundo, en el lado del usuario, por ejemplo en la industria, el gran número de puntos a modelar requiere que el LMFS tenga por un lado, un nivel elevado de autonomía para generar de la manera más desatendida posible los modelos. Por otro lado, necesita un nivel elevado de adaptabilidad para que, usando la misma estructura o metodología, pueda modelar diferentes tipos de cargas cuya procedencia pude variar significativamente. Por lo tanto, los sistemas de modelado abordados en esta tesis son aquellos que no solo buscan mejorar la precisión, sino también la adaptabilidad y autonomía. En busca de estos objetivos y soportados principalmente por algoritmos de inteligencia computacional, procesamiento de señales y estadística, hemos propuesto tres algoritmos novedosos para el desarrollo de un LMFS en el lado del usuario. El primero de ellos busca mejorar la adaptabilidad del LMFS manteniendo una buena precisión y capacidad de autonomía. Denominado ETA, consiste del uso de una estructura ANFIS que es entrenada por un algoritmo genético multi objetivo (MOGA). Como resultado de este híbrido, obtenemos un algoritmo con excelentes capacidades de generalización y fácil de adaptar para el entrenamiento y evaluación de nuevas estructuras adaptativas basadas en ANFIS. El segundo de los algoritmos desarrollados aborda la autonomía del LMFS para así poder generar modelos de múltiples cargas. Al igual que en la anterior propuesta usamos un ANFIS y un MOGA, pero esta vez el MOGA en vez de entrenar el ANFIS, se utiliza para encontrar la configuración cuasi-óptima del ANFIS. Encontrar la configuración apropiada de un ANFIS es muy importante para obtener un buen funcionamiento del LMFS en lo que a precisión y generalización respecta. El LMFS propuesto, además de configurar automáticamente el ANFIS, incluyó diversos algoritmos para procesar los datos puros que casi siempre estuvieron contaminados de datos espurios y gaps de información, operando satisfactoriamente en las condiciones de prueba en un escenario real. El tercero y último de los algoritmos buscó mejorar la precisión manteniendo la autonomía y adaptabilidad, aprovechando para ello la existencia de patrones dominantes de más baja resolución temporal que el consumo objetivo, y que son más fáciles de modelar y pronosticar. La metodología desarrollada se basa en la transformada de Hilbert-Huang para detectar y seleccionar tales patrones dominantes. Además, esta metodología define el uso de modelos parciales de los patrones dominantes seleccionados, para mejorar la precisión del LMFS y mitigar el problema de aleatoriedad que afecta a los consumos en el lado del usuario. Adicionalmente, se incorporó el algoritmo de auto configuración que se presentó en la propuesta anterior para hallar la configuración cuasi-óptima de los modelos parciales. Este punto fue crucial puesto que a mejor configuración de los modelos parciales mayor es la mejora en precisión del pronóstico final. Finalmente y para cerrar este trabajo de tesis, se realizó una prospección de las tendencias en cuanto al uso de iEMS y se esbozaron varias propuestas de algoritmos híbridos, cuyo estudio y comprobación se plantea en futuros estudios.

Sota una constant generació de nous models de negoci, les empreses necessiten informació per a la presa de decisions que els permeti aconseguir una posició mes competitiva al mercat. La presa de decisions en grups de treball col·laboratiu requereix del coneixement de l'estat actual dels temes importants i que afecten al grup, així com la trajectòria que han de prendre. No obstant això, hi ha una deficiència d’eines de software per donar suport a la recol·lecció d'opinions en equips de treball, i també se'n troba manca d'eines que implementin nous models per a l'avaluació crítica, prioritzada i consensuada de la situació actual y desitjada en una empresa per la presa de decisions. Aquest treball de recerca proposa un nou model de generació de consens a l'interior de grups de treball a través d'una eina WEB-GDSS implementada com a part del projecte, i analitza la validesa del model i de l'eina. / Bajo una constante generación de nuevos modelos de negocio, las empresas necesitan información para la toma de decisiones que les permita alcanzar una posición más competitiva en el mercado. Para la toma de decisiones los grupos de trabajo colaborativo requieren conocimiento del estado actual de los temas que les importan y afectan, así como del rumbo que deben tomar al respecto. Sin embargo existe una deficiencia en herramientas de software para apoyar la recolección de opiniones en equipos de trabajo, por lo que también existe deficiencia de herramientas que implementen nuevos modelos para la evaluación crítica, priorizada y consensuada de la situación existente y deseable en una empresa para la toma de decisiones. Este trabajo propone un nuevo modelo de generación de consenso al interior grupos de trabajo a través de una herramienta WEB-GDSS implementada como parte del proyecto, y analiza la validez del modelo y de la herramienta. / Under a constant new business models generation, firms need information for decision-making that allow them to have a competitive market position. Decision-making in collaborative workgroups requires knowledge about current status of the issues that matter and affect the group, as well as the direction to be taken in this regard. However, there is a lack of software to support the collection of opinions in workgroups, so there is also a lack of tools that implement new models for critical, prioritized, and agreed evaluation on existing and desirable company situation, needed to decisionmaking. This paper proposes a model of gathering consensus opinion within workgroups through a WEB GDSS which is implemented as part of the project, and discusses the validity of the model and the software.

Large scale computing infrastructures need scalable and effi cient resource allocation mechanisms to ful l the requirements of its participants and applications while the whole system is regulated to work e ciently. Computational markets provide e fficient allocation mechanisms that aggregate information from multiple sources in large, dynamic and complex systems where there is not a single source with complete information. They have been proven to be successful in matching resource demand and resource supply in the presence of sel sh multi-objective and utility-optimizing users and sel sh pro t-optimizing providers. However, global infrastructure metrics which may not directly affect participants of the computational market still need to be addressed -a.k.a. economic externalities like load balancing or energy-efficiency. In this thesis, we point out the need to address these economic externalities, and we design and evaluate appropriate regulation mechanisms from di erent perspectives on top of existing economic models, to incorporate a wider range of objective metrics not considered otherwise. Our main contributions in this thesis are threefold; fi rst, we propose a taxation mechanism that addresses the resource congestion problem e ffectively improving the balance of load among resources when correlated economic preferences are present; second, we propose a game theoretic model with complete information to derive an algorithm to aid resource providers to scale up and down resource supply so energy-related costs can be reduced; and third, we relax our previous assumptions about complete information on the resource provider side and design an incentive-compatible mechanism to encourage users to truthfully report their resource requirements effectively assisting providers to make energy-eff cient allocations while providing a dynamic allocation mechanism to users. / Les infraestructures computacionals de gran escala necessiten mecanismes d’assignació de recursos escalables i eficients per complir amb els requisits computacionals de tots els seus participants, assegurant-se de que el sistema és regulat apropiadament per a que funcioni de manera efectiva. Els mercats computacionals són mecanismes d’assignació de recursos eficients que incorporen informació de diferents fonts considerant sistemes de gran escala, complexos i dinàmics on no existeix una única font que proveeixi informació completa de l'estat del sistema. Aquests mercats computacionals han demostrat ser exitosos per acomodar la demanda de recursos computacionals amb la seva oferta quan els seus participants son considerats estratègics des del punt de vist de teoria de jocs. Tot i això existeixen mètriques a nivell global sobre la infraestructura que no tenen per que influenciar els usuaris a priori de manera directa. Així doncs, aquestes externalitats econòmiques com poden ser el balanceig de càrrega o la eficiència energètica, conformen una línia d’investigació que cal explorar. En aquesta tesi, presentem i descrivim la problemàtica derivada d'aquestes externalitats econòmiques. Un cop establert el marc d’actuació, dissenyem i avaluem mecanismes de regulació apropiats basats en models econòmics existents per resoldre aquesta problemàtica des de diferents punts de vista per incorporar un ventall més ampli de mètriques objectiu que no havien estat considerades fins al moment. Les nostres contribucions principals tenen tres vessants: en primer lloc, proposem un mecanisme de regulació de tipus impositiu que tracta de mitigar l’aparició de recursos sobre-explotats que, efectivament, millora el balanceig de la càrrega de treball entre els recursos disponibles; en segon lloc, proposem un model teòric basat en teoria de jocs amb informació o completa que permet derivar un algorisme que facilita la tasca dels proveïdors de recursos per modi car a l'alça o a la baixa l'oferta de recursos per tal de reduir els costos relacionats amb el consum energètic; i en tercer lloc, relaxem la nostra assumpció prèvia sobre l’existència d’informació complerta per part del proveïdor de recursos i dissenyem un mecanisme basat en incentius per fomentar que els usuaris facin pública de manera verídica i explícita els seus requeriments computacionals, ajudant d'aquesta manera als proveïdors de recursos a fer assignacions eficients des del punt de vista energètic a la vegada que oferim un mecanisme l’assignació de recursos dinàmica als usuaris

Recently the community started looking into Hardware/Software (HW/SW) co-designed processors as potential solutions to move towards the less power consuming and the less complex designs. Unlike other solutions, they reduce the power and the complexity doing so called dynamic binary translation and optimization from a guest ISA to an internal host custom ISA. This thesis tries to answer the question on how to simulate this kind of architectures. For any kind of processor's architecture, the simulation is the common practice, because it is impossible to build several versions of hardware in order to try all alternatives. The simulation of HW/SW co-designed processors has a big issue in comparison with the simulation of traditional HW-only architectures. First of all, open source tools do not exist. Therefore researches many times assume that the software layer overhead, which is in charge for dynamic binary translation and optimization, is constant or ignored. In this thesis we show that such an assumption is not valid and that can lead to very inaccurate results. Therefore including the software layer in the simulation is a must. On the other side, the simulation is very slow in comparison to native execution, so the community spent a big effort on delivering accurate results in a reasonable amount of time. Therefore it is the common practice for HW-only processors that only parts of application stream, which are called samples, are simulated. Samples usually correspond to different phases in the application stream and usually they are no longer than a few million of instructions. In order to archive accurate starting state of each sample, microarchitectural structures are warmed-up for a few million instructions prior to samples instructions. Unfortunately, such a methodology cannot be directly applied for HW/SW co-designed processors. The warm-up for HW/SW co-designed processors needs to be 3-4 orders of magnitude longer than the warm-up needed for traditional HW-only processor, because the warm-up of software layer needs to be longer than the warm-up of hardware structures. To overcome such a problem, in this thesis we propose a novel warm-up technique specialized for HW/SW co-designed processors. Our solution reduces the simulation time by at least 65X with an average error of just 0.75\%. Such a trend is visible for different software and hardware configurations. The process used to determine simulation samples cannot be applied to HW/SW co-designed processors as well, because due to the software layer, samples show more dissimilarities than in the case of HW-only processors. Therefore we propose a novel algorithm that needs 3X less number of samples to achieve similar error like the state of the art algorithms. Again, such a trend is visible for different software and hardware configurations. / Els processadors co-dissenyats Hardware/Software (HW/SW co-designed processors) han estat proposats per l'acadèmia i la indústria com a solucions potencials per a fabricar processadors menys complexos i que consumeixen menys energia. A diferència d'altres alternatives, aquest tipus de processadors redueixen la complexitat i el consum d'energia aplicant traducció y optimització dinàmica de binaris des d'un repertori d'instruccions (instruction set architecture) extern cap a un repertori d'instruccions intern adaptat. Aquesta tesi intenta resoldre els reptes relacionats a la simulació d'aquest tipus d'arquitectures. La simulació és un procés comú en el disseny i desenvolupament de processadors ja que permet explorar diverses alternatives sense haver de fabricar el hardware per a cadascuna d'elles. La simulació de processadors co-dissenyats Hardware/Software és un procés més complex que la simulació de processadores tradicionals, purament hardware. Per exemple, no existeixen eines de simulació disponibles per a la comunitat. Per tant, els investigadors acostumen a assumir que la capa de software, que s'encarrega de la traducció i optimització de les aplicacions, no té un pes específic i, per tant, uns costos computacionals baixos o constants en el millor dels casos. En aquesta tesis demostrem que aquestes premisses són incorrectes i que els resultats amb aquestes acostumen a ser molt imprecisos. Una primera conclusió d'aquesta tesi doncs és que la simulació de la capa software és totalment necessària. A més a més, degut a que els processos de simulació són lents, s'han proposat tècniques de simulació que intenten obtenir resultats precisos en el menor temps possible. Una pràctica habitual és la simulació només de parts de les aplicacions, anomenades mostres, en el disseny de processadors convencionals, purament hardware. Aquestes mostres corresponen a diferents fases de les aplicacions i acostumen a ser de pocs milions d'instruccions. Per tal d'aconseguir un estat microarquitectònic acurat per a cadascuna de les mostres, s'acostumen a estressar aquestes estructures microarquitectòniques del simulador abans de començar a extreure resultats, procés anomenat "escalfament" (warm-up). Desafortunadament, aquesta metodologia no pot ser aplicada a processadors co-dissenyats Hardware/Software. L'"escalfament" de les estructures internes del simulador en el disseny de processadores co-dissenyats Hardware/Software són 3-4 ordres de magnitud més gran que el mateix procés d' "escalfament" en simulacions de processadors convencionals, ja que en els primers cal "escalfar" també les estructures i l'estat de la capa software. En aquesta tesi proposem tècniques de simulació basades en l' "escalfament" de les estructures que redueixen el temps de simulació en 65X amb un error mig del 0,75%. Aquests resultats són extrapolables a diferents configuracions del hardware i de la capa software. Finalment, les tècniques convencionals de selecció de mostres d'aplicacions a simular no són aplicables tampoc a la simulació de processadors co-dissenyats Hardware/Software degut a que les mostres es comporten de manera molt diferent quan es té en compte la capa software. En aquesta tesi, proposem un nou algorisme que redueix 3X el nombre de mostres a simular comparat amb els algorismes tradicionals per a processadors convencionals per a obtenir un error similar. Aquests resultats també són extrapolables a diferents configuracions de hardware i de software. En conclusió, en aquesta tesi es respon al repte de com simular processadors co-dissenyats Hardware/Software, que són una alternativa al disseny tradicional de processadors. Hem demostrat que cal simular la capa software i s'han proposat noves tècniques i algorismes eficients d' "escalfament" i selecció de mostres que són tolerants a diferents configuracions