REAL-TIME SYSTEMS

Author: RUIZ MARTINEZ JOSE FRANCISCO.
Year: 2003.
University: POLITÉCNICA DE MADRID.
Place of defense: ESCUELA TECNICA SUPERIOR DE INGENIEROS DE TELECOMUNICACIÓN.
Place of preparation: ESCUELA TECNICA SUPERIOR DE INGENIEROS DE TELECOMUNICACION.

Summary: The objective of this work is to develop a hierarchical and decentralized architecture for the management of the quality of service and resources in embedded multimedia systems. The control system is hierarchical and multi provided, being able to manage the various entities conceptual properly. Work of this thesis provides architectural support needed to implement agents domains of applications capable of escapsular knowledge específicao of services . We have defined two different models that bear the covivencia between the manager and various Central specialized managers. Ul hand, is the model of partnership that control actions are implemented by the central manager using a single database containing the information needed to manage applications. Moreover, the delegation model provides a more flexible framework based on the specific functions delegated control agents specialized domains of applications. The validation of the architecture has been based on the development of a prototype funcional.La execution of a series of test cases has revealed the opreración system control.Se has also made a multiprocessor implementation of the control system that improves efficiency , robustness predictability and extensibility of the arqutectúra, distributing the burden of management between heterogeneous processors.

Author: CRUZ MARTIN ANA M..
Year: 2004.
University: MÁLAGA.
Place of defense: INGENIEROS INDUSTRIALES.
Place of preparation: E.T.S.I INDUSTRIALES.

Summary: This thesis deals with the problem of path planning in robotic systems composed of multiple elements. As a starting point, it takes an algorithm speeds planning for a single robot that generates trajectory whose profiles speed takes into account the speed limits, both physical and operational impact significantly on the movement of the vehicle. This algorithm monorrobot extends to the case multirobot providing a safe path to each of the vehicles comprising the system. Planning paths multirobot raised supports various solutions, whose degree of goodness is also different. Therefore, solving the problem involves not conveniently find any solution, but determining the best solution according to certain criteria can be established for each system multirobot. This paper analyzes in detail what type of tool would be best suited to determine the organization to provide optimal management concluded that genetic algorithms are a good choice, and also introduced a genetic algorithm that establishes the order of priorities that applies the algorithm planning multirobot trajectories. To facilitate the tasks of experimentation and testing has developed a graphical simulation systems multirobot, which avoids the need to employ a system multirobot real, which is usually very costly and complex to organize and manage. To this must be added a detailed compilation of the different architectures multirobot existing systems in the literature to date, as well as the study, rather broad, and the subsequent ranking of navigation deliberative strategies that can be found at present for this type systems.

Author: CALVO GORDILLO ISIDRO.
Year: 2004.
University: PAÍS VASCO.
Place of defense: E.T.S. DE INGENIEROS.
Place of preparation: ESCUELA TECNICA SUPERIOR DE INGENIEROS.

Summary: The systems of the modern industrial environments require constant adaptation to the new challenges posed by industry. In this vein, the possibility of remotely interact with these control systems are exciting benefits to different types of personnel involved. Some of these benefits are to reduce displacement to plant, reduce fault resolution times, or enforcement of certain processes in hazardous environments remotely. Moreover, it introduces the possibility of offering new services such as remote maintenance of the systems or the integration of information processing in the supply chains for customers and suppliers. Precisely, this paper proposes a design methodology that facilitates systematic and general building systems remote access to industrial plants. The methodology proposed for use of technologies that are seen as ripe for use in industrial environments having been used extensively in other settings. As part of the methodology, it is proposed an architecture framework consisting of various types of components proposed an architecture framework consisting of various types of components that must be completed in accordance with the guidelines set up the system for remote access. This methodology has been applied to two case studies distinct to test its validity. One such case study consists of a flexible manufacturing cell formed laboratory devices for industrial real, the latter being deployed in the electricity sector.

Author: USAMENTIAGA FERNANDEZ RUBEN.
Year: 2004.
University: OVIEDO.
Place of defense: DEPARTAMENTO DE INFORMATICA.
Place of preparation: DEPARTAMENTO DE INFORMATICA.

Author: JORDÀ PUIG SERGI.
Year: 2004.
University: POMPEU FABRA.
Place of defense: DEPARTAMENTO DE TECNOLOGÍA.
Place of preparation: UNIVERSIDAD POMPEU FABRA.

Summary: This thesis explores the use of the computer as a musical instrument for the implementation and improvisation of music in real time. Its main objective is to build a theoretical framework that allows us to assess the potential, the potential and diversity of these new digital musical instruments. The musical interactive systems dating back to the late sixties, when it developed the first computer-controlled analog synthesizers for use in concerts and interactive installations. Unfortunately, nearly four decades after the first pioneers, the design of computer-based musical instruments and, more generally, musical improvisation using computers, are still immature multidisciplinary areas in which knowledge does not seem sedimentarse of incrementally or cumulatively, and the wheel is reinvented permanently. The design of new digital musical instruments is a very broad field, covering high technology areas (eg electronics and sensor technologies, techniques for synthesis and audio processing, computer programming, etc.). Disciplines and linked to the study of human behavior (eg psychology, physiology, ergonomics, interaction humano-computadora, etc..), with all the possible connections between them. Much of the existing research, seeks to address separate parts of the problem. This approach, an approach that is essential to any real progress in this field, this is also, as show below, clearly inadequate. Comprehensive studies, which take into account not only technological or ergonomic factors, but also psychological, philosophical, conceptual, musicológicos, historical, and above all, musical, but can not be totally systematic, are absolutely necessary. This thesis deals with the idea of a digital instrument as a whole, regardless of the components within it and its potential characteristics (eg ways of controlling this type of sound generated, and so on.). She explores the essence and strengths of these new digital tools, as well as new paradigms of interpretation music and the music that they can bring unprecedented. We leave it the assumption that, to develop new and better tools must necessarily explore three directions parallel, consisting of: Â identify the quintessence of the new digital tools, which are truly original elements that can contribute to the musical performance as you can redefinirla; Â identify weaknesses and the weaknesses of traditional instruments, which have their limitations or problems could be eliminated, improved or resolved; Â identify generic essential qualities of the traditional instruments; those properties that probably such implied and assumed remain semiocultas, but we should never forget or dismiss. The identification of these points, it is because the primary objective of this thesis. There is a complex interplay between the tasks to imagine, design and build computers musical, execute and improvise music with them, and analyze and understand the results. This relationship can only be understood as a permanent progressive rapprochement. This thesis is the result of fifteen years of experience as a luthier improviser. It can be seen as a theoretical or conceptual work, while pilot experiments although it is documented passed along d 8 and years or 9f0 decades. For this reason, the thesis is divided into three parts. The first part identifies the three key points progressively above. It does so gradually introducing the new possibilities of digital tools, as well as providing a detailed description of the current technical knowledge and frameworks both technical and theoretical and conceptual in which researchers and luthiers digital players. It also presents several taxonomies that will help us develop a clear and concise picture of the scope of work. This part concludes in chapter seven, which introduced the first major contribution of this thesis, the development of a theoretical framework that allows us to assess the expressive possibilities that these new musical instruments can provide to their interpreters. The second part describes in detail seven of the musical instruments that I have developed over the years. Since the seven implementations are conceptually very different, each of which illustrates several of the paradigms introduced in the first part. Presented chronologically, these examples also illustrate the path that leads to the conceptual framework previously entered. The third incorporates the lessons and conclusions resulting from this evolutionary journey and introduces the second most important point of this thesis, suggesting possible solutions to achieve the goals set at the end of Part One. It concludes with what could be considered as my decalogue of lutheria digital, summarizes most of the ideas introduced in the argument, after which the reacTable *, the first instrument designed from scratch from all the concepts introduced in it, presented as future work.

Author: BALDASSARRI SANDRA SILVIA.
Year: 2004.
University: ZARAGOZA.
Place of defense: CENTRO POLITÉCNICO SUPERIOR.
Place of preparation: CENTRO POLITÉCNICO SUPERIOR.

Summary: The human being is characterized by proactively interact with the environment, so that move, speak or manipulate objects are common actions in our lives. It is only natural, then, the trend in the scientific world by recreating the verisimilitude with functional actions and behavior of human beings. The simulation of human beings have complex problems associated primarily due to the capacity of observer to detect any anomalies or irregularities in something as familiar as is the movement or body of a person. Despite advances in computer graphics in recent years in the field of animation of the human figure, it should be borne in mind that body shape can vary during animation, as it moves muscles, stretch or contracted and changing the external appearance. Therefore, to represent the complexity of the human body is essential to deal with the simulation of the muscles and strains that occur when it makes a move. This paper develops a complex system called MOBiL (Muscle defOrmation in Biped Locomotion), which allows encourage, in a coordinated manner, the global movement of the body and the local and volumetric strain of the muscles during locomotion. To do so, using a model based on two phases: skeletal and músculo-esqueletal. In Phase skeletal, realistic motion is obtained through a hybrid model of locomotion: dinámico-cinemático. The dynamic model is very simple and is the basis for a subsequent kinematic calculation which allows real-time simulation of the global movement for locomotion. In Phase músculo-esqueletal, deformation of the muscles are simulated physically with a system based on finite element, fueled by forces from the locomotion system through a model lines muscle action. The system works on a parametric generating animations of individuals with different anthropometric characteristics (height, weight and complexion) and to generate different locomociones (speed, length and frequency of the step).

Author: VICENTE BORRUEL LUIS.
Year: 2004.
University: ZARAGOZA.
Place of defense: CENTRO POLITECNICO SUPERIOR.
Place of preparation: CENTRO POLITECNICO SUPERIOR.

Summary: This dissertation deals with the study of algorithms adaptive signal processing applied in the field of active control, be it noise, vibration and acoustic structure. The algorithm FxLMS is the most widely used active control because of its simplicity yrobustez. However, convergence is often too slow, in addition to having a high dependence on the signal dereferencia. Thus, this thesis is focused on search algorithms altemativos for active control systems, with the objetivoprincipal to improve the characteristics of convergence of the algorithm FxLMS, and more specifically its speed. In addition delbeneficio obvious convergence faster, highly non-stationary environments that also translates into a mayorcapacidad tracking, and therefore cancellation. To keep the inevitable increase in the computational cost within deunos limits acceptable, as well as avoiding potential problems of numerical stability, only considered algorithms developed adaptive degradiente estocástico.Los algorithms are grouped into two families: the FxGAL, using an adaptive predictor in lattice for ortogonalizarel vector of input data, and W + FxLMS or FxLMS with preblanqueado, where a system bleach, transverse or encelosía determines the filtered signal reference. The performance of these algorithms are evaluated through exhaustive tests enentomos simulated, as well as in a real application of active control of noise in an environment of laboratorio.Además development and testing of new algorithms for active control in this thesis have been Two novedososestudios theorists about the convergence of adaptive algorithms. These analysis are, on the one hand, while deadaptación maximum FxLMS with deterministic reference signal, and second, to the influence that the desired signal takes laconvergencia algorithms descent gradient, especially when part of the initialization zero weights. Elprimero of them is a justification for the slow FxLMS. The second explains the cause of the improvement in speed deconvergencia in systems with preblanqueado.

Author: ENTRIALGO CASTAÑO JOAQUIN.
Year: 2005.
University: OVIEDO.
Place of defense: DEPARTAMENTO DE INFORMATICA. GIJON.
Place of preparation: DEPARTAMENTO DE INFORMATICA.

Summary: In real-time systems is of particular importance temporary correction, so that techniques have been developed that attempt to ensure that the requirements temporary. These techniques include the analysis methods planificabilidad, which determine from a model system if it is possible to obtain the results of the system within the time allotted. Moreover, we have developed systems that allow take time monitoring system implemented. The analysis techniques planificabilidad and monitoring tools often have no relationship between them, which complicates debugging errors behavior temporarily. This thesis presents a method that facilitates debugging errors behavioral models linking temporary employees during the analysis planificabilidad with measurements obtained from the system. The method is applied to two types of models: on the one hand, a "traditional" which is based on calculating the worst response time of the tasks of the system from its worst runtime, as well as a probabilistic model, , which calculates the probability of breaching the maximum terms of the tasks of the system from the probability distribution of their execution times. The method proposed in this thesis provides automatic search of errors behavior temporary, based on a comparison between the values of the model and measurements. This comparison is complicated in systems analyzed with probabilistic models, and we must compare two probability distributions. To make this comparison, we propose two metric heuristic, called "pessimism" and "optimism", which are the first approach to a complex problem is not addressed so far in the research of real-time systems. In order to test the applicability of the method has been implemented a debugging tool behavior temporary POSIX systems, as this standard the most important in the real-time operating systems today. The tool consists of two components: a monitoring system and a tool for analysis and visualization. The monitoring tool pays special importance to get a portable system instrumentation, so that work on POSIX systems regardless of their implementation and without the need for source code operating system. Further studies the disturbance introduced by the monitoring tool in the system measured, both from the standpoint of execution time consumed and the memory occupied. The analysis and visualization tool performs analysis based on the model and measurements, and presents the results in a number of windows that allow the user to analyze the causes of the error. Chief among these windows is the "window metric", which summarizes the most important results of the analysis. It highlights the errors, which can be analyzed in more detail through other windows Gantt diagrams that show the evolution of the system. When using probabilistic models, the tool also provides graphs to visually compare the probability distributions. Finally, we develop several case studies that allow on the one hand, see the advantages of the techniques presented in this thesis and, secondly, to analyze how the metric of pessimism and optimism.

Author: LOPEZ PEREZ BENJAMIN.
Year: 2005.
University: OVIEDO.
Place of defense: E.U.I.TEC. INFORMATICA.
Place of preparation: DEPARTAMENTO DE INFORMATICA.

Summary: The principle of separation of powers or responsibilities focuses on the ability module different parts of an application to a relevant concept, goal, task or specific purpose. An appropriate separation of the various aspects of a system reduces the complexity of the software, improving their understanding and maintainability, and facilitates code reuse. Considering the persistence as a matter typical of the majority of applications, this separation of the code's main system enables the development of programs without taking into account their requirements persistent, adding and adapting them downstream. This separation allows the developer to handle the persistence of programs independently to its functionality, as well as developing the business logic of the application without taking into account their requirements persistent. The persistence would, in that case, competition orthogonal that could be reused for any application, regardless of its functional requirements. After analyzing the various alternatives with the objective of achieving a complete separation from the appearance of persistence, it is clear that while some offer more transparent than others, allows any application to develop a code which is absolutely independent of the aspect of the characteristics of persistence . Similarly, most of the existing alternatives are dependent on a specific language. In addition to the deficiencies identified in the separation aspect of persistence, the existing approaches do not address allocation, separation or adapt at runtime. An adaptation dynamics of the various parameters related to the persistence of a system, as well as its allocation and disposal at runtime, are relevant and adaptive systems adaptable to emerging contexts time programming. The reflectivity computing is a technique that allows to adapt the structure and behavior of a system at runtime. About a platform reflective unrestricted offering computational reflectivity at runtime, has developed a system of persistence which demonstrates how this mechanism can be employed to provide a total separation of the business of persistence in a manner independent of the language programming selected. Additionally, the reflectivity computational offers a high level of adaptability to change dynamically features persistence of a program running on a program.

Author: BELDA ORTEGA ÁNGELA.
Year: 2005.
University: POLITÉCNICA DE VALENCIA.
Place of preparation: UNIVERSIDAD POLITÉCNICA DE VALENCIA.

Summary: The transmission of video streams on heterogeneous channels, which is the last link wireless presents interesting challenges for researchers engaged in the issue. Most of these systems have a very limited bandwidth. Besides the loss of packets in the fixed links are of a different nature to losses Package wireless links. Other aspects to consider are the existence of other flows competing for the available bandwidth and the variability of the wireless link. The video encoding, tools to protect against errors, transmission protocols and adapting to the properties snapshots of the channel are the main areas of research in which it seeks to improve the quality, both objective and subjective, stream video received . The popularity recently acquired by the mobile terminal devices (PDAs, mobile phones) has added a new degree of difficulty to efforts to move applications for transmitting video streams to these devices. This is due to the limited capacity of processing and memory space program available at the terminals. The work done in this thesis focuses on the study and analysis of the most common problems encountered in the development of a multimedia system for the transmission of video and audio streams in environments with low bandwidth and / or high rate of losses mobile devices. In this study have been analyzed, evaluated and implemented solutions to these problems. Beginning with the selection of an encoder suitable for such a system, has made an assessment of the options available for encoding can know the answer to this encoder. Then the encoder has been integrated into a multimedia middleware that has been used as a platform for the development of multimedia applications adaptive. This multimedia middleware has been analyzed, and improved ported to various mobile devices. Finally arising problems in multimedia communications with mobile terminals analyzing algorithms used to solve them, and proposes an algorithm and evaluate protective adaptive channel wireless based on the information feedback provided by the receiver.

Author: ÁLVAREZ PALOMO JOSÉ MARÍA.
Year: 2005.
University: MÁLAGA.
Place of defense: ESCUELA TÉCNICA SUPERIOR DE INGENIERÍA INFORMÁTICA.
Place of preparation: ESCUELA TÉCNICA SUPERIOR DE INFORMÁTICA.

Summary: This paper proposes a methodology for the development of real-time systems makes special emphasis on the consideration of non-functional requirements characteristic of this type of system requirements as temporary, concurrency, prioritizing or interaction with physical devices . The methodology takes elements from other existing, as SOMT and OCTOPUS and proposed mechanisms to solve problems as part of the passage of model objects to the process model and prioritization. The methodology is divided into four phases divided into two distinct areas, and the functional aspects of the non-functional. Throughout the methodology used and goal orientation UML. To take advantage of formal methods, such as simulation, validation and code generation proposes a formal semantics for some of the dynamic aspects of UML, specifically the actions and state machines. The semantics proposal in metamodelado and is based on the language MML. They distinguish between the abstract syntax and semantic domain. The valid elements of both sets are defined by class diagrams, which will be valid instances, and restrictions expressed in the functional language OCL. The elements of both sets are related to one another through the semantics, which implies a ratio of one (in the abstract syntax, the end "OF") many (in the semantic domain, the extreme "instances"). With this scheme, it has been defined semantics for actions and executions, with a hierarchy of classes for different kinds of actions and executions, in the first level of this hierarchy are distinguished primitive actions and compound. An action is defined as a computer that modifies the state of an element of the system and the performances are individual occurrences of the shares. It has also been defined with the same semantics strategy for the state machines, which are considered the most of the features included in the form of UML, as states compounds, sequential and concurrent actions input and output states and guaridas transitions and events. In this case, we have followed a strategy that increases in part of a simple model and successive refinements are adding new features.

Author: GUITARTE PEREZ JESUS FERNANDO.
Year: 2005.
University: ZARAGOZA.
Place of defense: CENTRO POLITECNICO SUPERIOR.
Place of preparation: CENTRO POLITECNICO SUPERIOR.

Summary: In this thesis the visual information contained in the movement of the lips has been used to improve the robustness against noise in voice recognition systems on devices with limited resources. The system described herein significantly reduces the error rate in high acoustic noise levels. The algorithms used are characterized by low consumption, both in processing time and memory, allowing its use in integrated devices. The main issues to take into consideration in a system of reading lips are tracing and tracking of the lips, the extraction of visual information and its integration with acoustic information. In this paper we propose solutions to these three problems suitable for use in devices with limited resources. It has developed an algorithm for locating and tracking lips. Based on a ranking by color, using horizontal contours and a simple model of the face algorithm implemented provides the position of the mouth with a very low consumption of resources. This algorithm has been implemented in a mobile phone processing rate of 15 frames per second in real time. In addition to the extraction of visual information have studied two types of different algorithms, one based on a modeling geometry lip and another based on a mathematical transformation of pixels in the region of the mouth. It has been shown in devices with limited resources the second type provides better rates of appreciation by not requiring the removal accurate contour of the lips. Finally, we have studied three techniques for integrating sound and visual information, which differ in the position where the integration in the process of recognition: early and late hybrid. It has been found that the last provides the best results for recognition, presenting further advantages for its implementation in an integrated device. The system design provides very good results in the fight against the harmful effects of noise and interference from another speaker, relative reductions of 38% error rate. The conventional techniques to reduce noise, such as theft or spectral filtering Wiener, fail to reduce the error rate compared to this type of interference. Taking into account the increase in the growing market for devices equipped with camera and the viability of this technology demonstrated in this thesis, reading lips can be seen as a technology to provide more robustness to future systems for voice recognition.

Author: SANZ MONTEMAYOR ANTONIO.
Year: 2005.
University: REY JUAN CARLOS.
Place of defense: ESCUELA SUPERIOR DE CIENCIAS EXPERIMENTALES Y TECNOLOGÍA.
Place of preparation: ESCUELA SUPERIOR DE CIENCIAS EXPERIMENTALES Y TECNOLOGÍA.

Summary: The high demand for the industry of videosjuegos has facilitated the growth of graphics hardware to become a platform for high performance and low cost. Beyond rendering tasks, graphics will allow processing data in a very efficient parallel if we have the algorithms and data structures suitable. In this thesis, presented new proposals algorithmic adapted to a graphical platform, which has been validated through various proposals relating to image processing to eventually generate different contributions to algoritmicas in trouble processing video. The format interlaced video, as a mechanism for reducicón bandwidth for the transmission of the signal, had its origins during the early years of the introduction of television. We have proposed wide variety of algorithms for reconstruction or deinterlacing. However, few of them have adapted soft computing techniques, in particular the muddy logic to this problem. In this thesis, improving a preliminary proposal for deinterlacing fuzzy then adapt the model to a platform graphical consumption. Moreover, the tracking of objects in image sequences is a problem on the computer vision that has many applications in various tasks of knowledge. The particulate filters belong to a class of algorithms for estimating sequential specializing in troubleshooting from a dynamic perspective Bayes. In this thesis, are modeled the foundations of a particulate filter applied to the visual tracking to be executed in a graphics platform, adapting their computational model to the traditional scheme of computation of a graphics card, adapting their computational model to the traditional scheme of computation of a graphics card in common use.

Author: OCAMPO MARTINEZ CARLOS AUGUSTO.
Year: 2006.
University: POLITÉCNICA DE CATALUÑA.
Place of defense: E.U. Eng. Tècnica Industrial Terrassa.
Place of preparation: EDIFICI U DESPATX 518 Campus SUD.

Summary: Real time control (RTC) of sewer networks plays a fundamental role in the management of hydrological systems, both in the urban water cycle, as well as in the natural water cycle. An adequate design of control systems for sewer networks can prevent the negative impact on the environment that Combined Sewer Overflow (CSO) as well as preventing flooding within city limits when extreme weather conditions occur. However, sewer networks are large scale systems with many variables, complex dynamics and strong nonlinear behaviour. Any control strategy applied should be capable of handling these challenging requirements. Within the field of RTC of sewer networks for global network control, the Model Predictive Control (MPC) strategy stands out due to its ability to handle large scale, nonlinear and multivariable systems. Furthermore, this strategy allows performance optimization, taking into account several control objectives simultaneously. This thesis is devoted to the design of MPC controllers for sewer networks, as well as the complementary modelling methodologies. Furthermore, scenarios where actuator faults occur are specially considered and strategies to maintain performance or at least minimizing its degradation in presence of faults are proposed. In the first part of this thesis, the basic concepts are introduced: sewer networks, MPC and fault tolerant control. In addition, the modelling methodologies used to describe such systems are presented. Finally the case study of this thesis is described: the sewer network of the city of Barcelona (Spain). The second part of this thesis is centered on the design of MPC controllers for the proposed case study. Two types of models are considered: (i) a linear model whose corresponding MPC strategy is known for its advantages such as convexity of the optimization problem and existing proofs of stability, and (ii) a hybrid model which allows the inclusion of state dependent hybrid dynamics such as weirs. In the latter case, a new hybrid modelling methodology is introduced and hybrid model predictive control (HMPC) strategies based on these models are designed. Furthermore, strategies to relax the optimization problem are introduced to reduce calculation time required for the HMPC control law. Finally, the third part of this thesis is devoted to study the fault tolerance capabilities of MPC controllers. Actuator faults in retention and redirection gates are considered. Additionally, hybrid modelling techniques are presented for faults which, in the linear case, can not be treated without loosing convexity of the related optimization problem. Two fault tolerant HMPC strategies are compared: the active strategy, which uses the information from a diagnosis system to maintain control performance, and the passive strategy which only relies on the intrinsic robustness of the MPC control law. As an extension to the study of fault tolerance, the admissibility of faulty actuator configurations is analyzed with regard to the degradation of control objectives. The method, which is based on constraint satisfaction, allows the admissibility evaluation of actuator fault configurations, which avoids the process of solving the optimization problem with its related high computational cost.

Author: Guirado Fernández Fernando.
Year: 2006.
University: AUTÓNOMA DE BARCELONA.
Place of defense: Escuela Técnica Superior de Ingeniería.
Place of preparation: Escuela Técnica Superior (ETSE).

Summary: Applications of parallel processing continuous streams of input data are of great interest to the scientific community. For these applications to optimize their performance based on two different criteria. The primreo corresponds to the reduction of latency, so as to obtain the result for data entry in the shortest possible time. The second is defined on the basis of a predetermined ratio of data processed per unit time, value named as productivity. The need to process a continuous stream of data makes these applications add a factor of iteratividad in carrying out its mission, which represents an increase of complexity in regard to its optimization to be done in parallel computing in that each run only processed data. This feature coupled with the structure and dependencies between tasks available in stages, makes these applications are named as applications pipeline. The aim of this thesis work has been defined to provide a solution for the optimization of pipeline applications. In the process of optimization submitted, addresses both the definition of the structure of the graph tasks for those situations in which the characteristics of the application is not possible to achieve the requirement for productivity marked, as the manner in which the tasks of application must be allocated to the architecture on which to run, ensuring that remains the performance targets. The architecture chosen as a base for carrying out these tasks, it is incumbent upon the architecture of distributed memory, as this is the most widely used today in the high-performance computing. For a definition of tasks graph has developed two techniques, which, under the approach to minimize communications, can obtain for the graph structure best suited to meet the requirement of productivity marked as a target. Both techniques are located tasks that act as bottleneck of the application and according to their characteristics and the type of data stream to deal applies to: (a) parallel data to reduce the computational time of each of , or (b) replication tasks to increase capacity to process, in parallel, more data input stream. For the process of assigning tasks on the processing nodes, it has developed two heuristics of mapping based on the concept of phase synchronous. These heuristics have been labeled as MPASS (Pipeline Mapping of Applications based on Synchronous Stages) and MPART (Pipeline Mapping of Applications based on Reduced Tree). Both heuristics, took aim of allocating two possible options: (a) Obtain a pre productivity, minimizing the number of nodes or processing (b) minimizing latency, under a requirement to achieve productivity and minimizing the number of nodes processing. The experiment was carried out using synthetic applications that shape the behavior of the applications pipeline and three real-world applications of different scientific fields: the compressor video MPEG2, IVUS (IntraVascular UltraSound), which was responsible for processing medical images and BASIZ (Bright And Satured Images Zones), which detects a sequence of images, those regions that capture the attention of the human eye. The results demonstrate how the techniques proposed are able to improve the performance of applications, from defining a new dependency of tasks to improve performance, as well as allocating more suitable for this type of applications.