Search

A complex and important task in the cloud resource management is the efficient allocation of virtual machines (VMs), or containers, in physical machines (PMs). The evaluation of VM placement techniques in real-world clouds can be tedious, complex and time-consuming. This situation has motivated an increasing use of cloud simulators that facilitate this type of evaluations. However, most of the reported VM placement techniques based on simulations have been evaluated taking into account one specific cloud resource (e.g., CPU), whereas values often unrealistic are assumed for other resources (e.g., RAM, awaiting times, application workloads, etc.). This situation generates uncertainty, discouraging their implementations in real-world clouds. This paper introduces CloudBench, a methodology to facilitate the evaluation and deployment of VM placement strategies in private clouds. CloudBench considers the integration of a cloud simulator with a real-world private cloud. Two main tools were developed to support this methodology, a specialized multi-resource cloud simulator (CloudBalanSim), which is in charge of evaluating VM placement techniques, and a distributed resource manager (Balancer), which deploys and tests in a real-world private cloud the best VM placement configurations that satisfied user requirements defined in the simulator. Both tools generate feedback information, from the evaluation scenarios and their obtained results, which is used as a learning asset to carry out intelligent and faster evaluations. The experiments implemented with the CloudBench methodology showed encouraging results as a new strategy to evaluate and deploy VM placement algorithms in the cloud. This work was partially funded by the Spanish Ministry of Economy, Industry and Competitiveness under the Grant TIN2016-79637-P “Towards Unifcation of HPC and Big Data Paradigms” and by the Mexican Council of Science and Technology (CONACYT) through a Ph.D. Grant (No. 212677).

The transmission of COVID-19 through a population depends on many factors which model, incorporate, and integrate a large number of heterogeneous data sources. The work we describe in this paper focuses on the data management aspect of EpiGraph, a scalable agent-based virus-propagation simulator. We describe the data acquisition and pre-processing tasks that are necessary to map the data to the different models implemented in EpiGraph in a way that is efficient and comprehensible. We also report on post-processing, analysis, and visualization of the outputs, tasks that are fundamental to make the simulation results useful for the final users. Our simulator captures complex interactions between social processes, virus characteristics, travel patterns, climate, vaccination, and non-pharmaceutical interventions. We end by demonstrating the entire pipeline with one evaluation for Spain for the third COVID wave starting on December 27th of 2020. This work has been supported by the Spanish Instituto de Salud Carlos III under the project grant 2020/00183/001, the project grant BCV-2021-1-0011, of the Spanish Supercomputing Network (RES) and the European Union's Horizon 2020 JTI-EuroHPC research and innovation program under grant agreement No 956748.

XXXVIII Congreso Anual de la Sociedad Española de Ingeniería Biomédica. 25 – 27 Nov, 2020 En sistemas de tomografía axial computarizada (TAC) es común hacer varias adquisiciones tomográficas consecutivas para distintas posiciones de la cama, que posteriormente se combinan para ampliar el campo de visión en la dirección longitudinal. Para esta combinación es necesaria la calibración geométrica del movimiento de la cama para evitar dobles bordes en la zona de pegado. Esta calibración se realiza de forma periódica usando un maniquí específico. Este trabajo presenta una herramienta novedosa de pegado de cama automático para TAC basada en correlación. Nuestra propuesta explota el paralelismo masivo que ofrecen las GPUs y cuenta con un modelo de memoria optimizado que permite pegar grandes volúmenes en tiempo casi real. La evaluación en estudios de roedor demuestra, no solo que la implementación ofrecida es capaz de pegar los estudios tomográficos en un tiempo reducido, sino también que hace un uso eficiente de los recursos de memoria disponibles. Este trabajo ha sido financiado por el Ministerio de Ciencia, Innovación y Universidades (Instituto de Salud Carlos III, proyecto DTS17/00122; Agencia Estatal de Investigación, proyecto DPI2016-79075-R-AEI/FEDER, UE), cofinanciado por Fondos de la Unión Europea (FEDER), “A way of making Europe”. Además, ha sido financiado por el Programa de apoyo a la realización de proyectos interdisciplinares de I+D para jóvenes investigadores de la Universidad Carlos III de Madrid 2019-2020 en el marco del Convenio Plurianual Comunidad de Madrid- Universidad Carlos III de Madrid (proyecto DEEPCT-CM-UC3M) y por CRUE Universidades, CSIC y el Banco Santander (Fondo Supera, proyecto RADCOV19).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. As data-intensive scientific prevalence arises, there is a necessity of simplifying the development, deployment, and execution of complex data analysis applications. The Data Mining Cloud Framework is a service-oriented system for allowing users to design and execute data analysis applications, defined as workflows, on cloud platforms, relying on cloud-provided storage services for I/O operations. Hercules is an in-memory I/O solution that can be deployed as an alternative to cloud storage services, providing additional performance and flexibility features. This work extends the DMCF-Hercules cooperation by applying novel data placement and task scheduling techniques for exposing and exploiting data locality in data-intensive workflows. This work is partially supported by EU under the COST Program Action IC1305: Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. The evolution in scientific computing towards data-intensive applications and the increase of heterogeneity in the computing resources, are exposing new challenges in the I/O layer requirements. We propose a generic I/O architecture for data-intensive applications based on in-memory distributed caching. This solution leverages the evolution of network capacities and the price drop in memory to improve I/O performance for I/O-bounded applications adaptable to existing high-performance scenarios. We have showed the potential improvements European Cooperation in Science and Technology. COST This work is partially supported by EU under the COST Program Action IC1305: Network for Sustainable Ultrascale Computing (NESUS). This work is partially supported by the grant TIN2013-41350-P, Scalable Data Management Techniques for High-End Computing Systems from the Spanish Ministry of Economy and Competitiveness.

Technical report Govein E-Invoice, Co-financed by the European Union, Connecting Europe facilities. Agreement number INEA/CEF/ICT/A2015/1129678, Action No. 2015-EU-IA-0058

Las desventajas presentes en las plataformas de computación actuales y la fácil migración a la computación en la nube, han logrado que cada vez más aplicaciones científicas se adapten a los distintos frameworks de computación distribuida basadas en flujo de tareas. Sin embargo, muchas de ellas ya han sido optimizadas para su ejecución en aceleradores tales como GPUs. En este trabajo se presenta una arquitectura que facilita la ejecuión de aplicaciones tradicionalmente basadas en entornos HPC al nuevo paradigma de computación Big Data. Además, se demuestra cómo gracias a una mayor capacidad de memoria, el reparto automático de tareas y a la mayor potencia de cálculo de los sistemas heterogéneos se puede converger a un nuevo modelo de ejecución altamente distribuido. En ese trabajo se presenta un estudio de viabilidad de esta propueta mediante la utilización de GPUs dentro de la infraestructura de cómputo Spark. Esta arquitectura será evaluada a través de una aplicación de tratamiento de imagen médica. Los resultados demuestran que aunque nuestra arquitectura sobre un nodo no produce resultados absolutos mejores que la aplicación original, según se aumenta el número de GPUs y por lo tanto la ocupación de estas influye más la aplicación basada en Spark se acerca al rendimiento del simulador original. Finalmente, realizamos un estudio de la ocupación de las GPUs empleadas para las distintas políticas propuestas, demostrando que al tener en cuenta las características dinámicas de la GPUs (número de tareas en ejecución) podemos tener una mayor ganancia de rendimiento. Este trabajo ha sido parcialmente subvencionado por el Ministerio de Economia y Competitividad, bajo el proyecto TIN2013-41350-P, Scalable Data Management Techniques for High-End Computing Systems.

En muchos de los campos de la investi gación científica, se ba establecido Matlab como he rramienta de facto para el diseño de aplicaciones. Esta aproximación o&ece mucbaa ventajas como el rápido despliegue de prototipos, alto rendimiento en álge Lrta liu..ml, o:1uL1·o:1 uLrvb. Siu ,,u.uL~"~ hu. ~vli~1,;ivuo:1b desarrolladas son altamente dependientes del motor de ejecución de Matlab, limitando su despliegue en multitud de plataformas de altas prestaciones. En este trabajo presentamos un caso práctico de migración de una aplicación inicialmente basada en Matlab a una aplicación nativa en lenguaje e++. Pa ra ello se presentará la metodología empleada para la migración y las herramientas que facilitan esta tarea. La evaluación llevada a cabo demuestta que la solución implementada ofrece un buen rendimiento sobre dis tintas plataformas y sistemas altamente heterogéneo Este trabajo ha sido financiado por el Proyecto Europeo ICT 644235 RePhrase: REfactoring Parallel Heterogeneous Resource-Aware Applicationsy el Ministerio de Economia y Competitividad, bajo el proyecto TIN2013-41350-P Scalable Data Management Techniques for High-End Computing Systems

Proceedings of: Second International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2015). Krakow (Poland), September 10-11, 2015.

The ever-increasing power of supercomputer systems is both driving and enabling the emergence of new problem-solving methods that require the efficient execution of many concurrent and interacting tasks. Swift/T, as a description language and runtime, offers the dynamic creation and execution of workflows, varying in granularity, on high-component-count platforms. Swift/T takes advantage of the Asynchronous Dynamic Load Balancing (ADLB) library to dynamically distribute the tasks among the nodes. These tasks may share data using a parallel file system, an approach that could degrade performance as a result of interference with other applications and poor exploitation of data locality. The objective of this work is to expose and exploit data locality in Swift/T through Hercules, a distributed in-memory store based on Memcached, and to explore tradeoffs between data locality and load balance in distributed workflow executions. In this paper we present our approach to enable locality-based optimizations in Swift/T by guiding ADLB to schedule computation jobs in the nodes containing the required data. We also analyze the interaction between locality and load balance: our initial measurements based on various raw file access patterns show promising results. Moreover, we present future work based on the promising results achieved so far. This material is based upon work supported by the U.S. Department of Energy, Office of Science, under contract DE-AC02-06CH11357. Computing resources were provided by the Argonne Leadership Computing Facility. The work presented in this paper was supported by the COST Action IC1305, “Network for Sustainable Ultrascale Computing (NESUS).” The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement number 328582.

Cloud-based storage is becoming a cost-effective solution for agencies, hospitals, government instances and scientific centers to deliver and share contents to/with a set of end-users. However, reliability, privacy and lack of control are the main problems that arise when contracting content delivery services with a single cloud storage provider. This paper presents the implementation of a storage system for content delivery and sharing in federated cloud storage networks. This system virtualizes the storage resources of a set of organizations as a single federated system, which is in charge of the content storage. The architecture includes a metadata management layer to keep the content delivery control in-house and a storage synchronization worker/monitor to keep the state of storage resources in the federation as well as to send contents near to the end-users. It also includes a redundancy layer based on a multi-threaded engine that enables the system to withstand failures in the federated network. We developed a prototype based on this scheme as a proof of concept. The experimental evaluation shows the benefits of building content delivery systems in federated cloud environments, in terms of performance, reliability and profitability of the storage space. The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of: First International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2014). Porto (Portugal), August 27-28, 2014.

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. The PhD Symposium was a very good opportunity for the young researchers to share information and knowledge, to present their current research, and to discuss topics with other students in order to look for synergies and common research topics. The idea was very successful and the assessment made by the PhD Student was very good. It also helped to achieve one of the major goals of the NESUS Action: to establish an open European research network targeting sustainable solutions for ultrascale computing aiming at cross fertilization among HPC, large scale distributed systems, and big data management, training, contributing to glue disparate researchers working across different areas and provide a meeting ground for researchers in these separate areas to exchange ideas, to identify synergies, and to pursue common activities in research topics such as sustainable software solutions (applications and system software stack), data management, energy efficiency, and resilience. European Cooperation in Science and Technology. COST

Data movement is becoming a key issue in terms of performance and energy consumption in high performance computing (HPC) systems, in general, and Exascale systems, in particular. A preliminary step to perform I/O optimization and face the Exascale challenges is to deepen our understanding of energy consumption across the I/O stacks. In this paper, we analyze the power draw of different I/O operations using a new fine-grained internal wattmeter while simultaneously collecting system metrics. Based on correlations between the recorded metrics and the instantaneous internal power consumption, our methodology identifies the significant metrics with respect to power consumption and decides which ones should contribute directly or in a derivative manner. This approach has the advantage of building I/O power models based on a previous set of identified utilization metrics. This technique will be validated using write operations on an Intel Xeon Nehalem server system, as writes exhibit interesting patterns and distinct power regimes. The work presented in this paper has been partially supported by the EU Project FP7 318793 “EXA2GREEN” and partially supported by the EU under the COST Programme Action IC1305, “Network for Sustainable Ultrascale Computing (NESUS)” and by the grant TIN2013-41350-P, Scalable Data Management Techniques for High-End Computing Systems from the Spanish Ministry of Economy and Competitiveness. European Community's Seventh Framework Program

The Data Mining Cloud Framework (DMCF) is an environment for designing and executing data analysis workflows in cloud platforms. Currently, DMCF relies on the default storage of the public cloud provider for any I/O related operation. This implies that the I/O performance of DMCF is limited by the performance of the default storage. In this work we propose the usage of the Hercules system within DMCF as an ad-hoc storage system for temporary data produced inside workflow-based applications. Hercules is a distributed in-memory storage system highly scalable and easy to deploy. The proposed solution takes advantage of the scalability capabilities of Hercules to avoid the bandwidth limits of the default storage. Early experimental results are presented in this paper, they show promising performance, particularly for write operations, compared to the performance obtained using the default storage services. This work is partially supported by EU under the COST Program Action IC1305: Network for Sustainable Ultrascale Computing (NESUS). This work is partially supported by the grant TIN2013-41350-P, Scalable Data Management Techniques for High-End Computing Systems from the Spanish Ministry of Economy and Competitiveness.

Actas de: XXIX Congreso Anual de la Sociedad Espñaola de Ingeniería Biomédica (CASEIB 2011). Cáceres, 16-18 Noviembre 2011. La mayoría de los tomógrafos para pequeño animal están basados en geometría cone-beam con un detector plano orbitando en trayectoria circular. La reconstrucción en estos sistemas se suele hacer con un método basado en el algoritmo propuesto por Feldkamp, Davis y Kress (FDK). El aumento de velocidad en la reconstrucción para tomografía rayos X (TAC) es un requisito fundamental para la extensión de su aplicación clínica. En este artículo se presenta una implementación eficiente de unn algoritmo de reconstrucción modular basado en FDK, que aprovecha las posibilidades de cómputo paralelo y la eficiente interpolación provista en CUDA al usar memoria de texturas que ofrecen las unidades de procesamiento gráfico (GPU). El algoritmo implementado, probado en un micro-TAC de alta resolución, presenta una mejora de velocidad de ejecución de la etapa de retroproyección de un factor 40x respecto a una implementación secuencial de referencia escrita e C, manteniéndose en todo mmento la calidad de la reconstrucción. Este trabajo ha sido financiado parcialmente por los proyectos AMIT Projects del programa CDTI CENIT, TEC2007-64731, TEC2008-06715-C02-01, RD07/0014/2009, TRA2009 0175, RECAVA-RETIC, RD09/0077/00087 (Ministerio de Ciencia e Innovación), ARTEMIS S2009/DPI-1802 (Comunidad de Madrid) y TIN2010-16497 (Ministerio de Ciencia e Innovación). Publicado

El grupo de investigación ARCOS de la Universidad Carlos III de Madrid (España), lleva varios años trabajando en sistemas de almacenamiento de datos flexibles y adaptativos. Sus sistemas de almacenamiento se caracterizan porque se gobiernan mediante software, lo que permite implementarlos sobre distintas plataformas hardware asegurando su portabilidad, se adaptan dinámicamente a las circunstancias de los sistemas siguiendo el paradigma de los sistemas autónomos y permiten obtener partido de sistemas con almacenamiento de datos distribuidos o repartidos entre múltiples dispositivos. Una de las tecnologías del grupo es el sistema AFTSYS. Un sistema de almacenamiento tolerante a fallos a nivel de objetos persistentes distribuidos, configurable por el usuario y adaptable al comportamiento del sistema.

Research group ARCOS of Universidad Carlos III de Madrid (Spain) have been working on flexible and adaptive data storage systems for several years. The storage systems developed are featured by software governance, making them portable across different hardware storage resources, and their dynamic adaptativy to the different circumstances of computer systems following the autonomic system paradigm. They also allow getting high performance storage by using data distribution or striping across multiple devices. One of the group’s technologies y the AFTSYS system. A fault-tolerant storage system for persistent distributed objects, user configurable and adaptive to system behaviour.

Research group ARCOS of Universidad Carlos III de Madrid (Spain) have been working on wireless sensor networks and their application to data recording from several sources. As part of this research we have developed a platform to automatically create and deploy secure applications for wireless sensors networks using MDA (ADSA) to develop these applications intuitively by using a GUI, and by including new and existing components for WSN, such as secure communications, ciphering, etc. This technology is useful for every kind of application using WSN and to create and deploy sensor networks in a semiautomatic way. It has several advantages compared with the traditional development, very expensive and error prone. The platform itself introduces an enhanced security by reducing to a minimum development errors and by provided a controlled development platform. Moreover, it allows to include new or legacy components, such as secure communications, easily an intuitively.

El grupo de investigación ARCOS de la Universidad Carlos III de Madrid (España), lleva varios años trabajando en sistemas de almacenamiento de datos flexibles y adaptativos. Sus sistemas de almacenamiento se caracterizan porque se gobiernan mediante software, lo que permite implementarlos sobre distintas plataformas hardware asegurando su portabilidad, se adaptan dinámicamente a las circunstancias de los sistemas siguiendo el paradigma de los sistemas autónomos y permiten obtener partido de sistemas con almacenamiento de datos distribuidos o repartidos entre múltiples dispositivos. Una de las tecnologías del grupo es el sistema AFTSYS. Un sistema de almacenamiento tolerante a fallos a nivel de objetos persistentes distribuidos, configurable por el usuario y adaptable al comportamiento del sistema.

Ponencias de las Decimoséptimas Jornadas de Paralelismo de la Universidad de Castilla-La Mancha celebradas el 18,19 y 20 de septiembre de 2006 en Albacete Web servers scalability has been traditionally solved by improving software elements or increasing hardware resources of the server machine. Another approach has been the usage of distributed architectures. In such architectures, usually, file al- location strategy has been either full replication or full distribution. In previous works we have showed that partial replication offers a good balance between storage capacity and reliability. It offers much higher storage capacity while reliability may be kept at an equivalent level of that from fully replicated solutions. In this paper we present the architectural details of Web cluster solutions adapted to partial replication. We also show that partial replication does not imply a penalty in performance over classical fully replicated architectures. For evaluation purposes we have used a simulation model under the OMNeT++ framework and we use mean service time as a performance comparison metric. Publicado

The aim of this paper is to apply two adaptive control formulations under uncertainty, say open-loop and closed-loop, to the process of developing maintenance and repair policies for railway infrastructures. To establish the optimal maintenance and repair policies for railway lines, we use a previous design of risk model based on two factors: the criticality and the deterioration ratios of the facilities. Thus, our theory benefits from the Reliability Centered Management methodology application, but it also explicitly models uncertainty in characterizing a facility deterioration rate to decide the optimal policy to maintain the railway infrastructures. This may be the major contribution of this work. To verify the models presented, a computation study has been developed and tested for a real scenario: the railway line Villalba-Cercedilla in Madrid (Spain). Our results demonstrate again that applying any adaptive formulation, the cost of the railway lines maintenance shown is decreased. Moreover applying a Closed Loop Formulation the cost associated to the risk takes smaller values (40% less cost for the same risk than the deterministic approach), but with an Open Loop formulation the generated risk in the railway line is also smaller.

The aim of this paper is to apply two adaptive control formulations under uncertainty, say open-loop and closed-loop, to the process of developing maintenance and repair policies for railway infrastructures. To establish the optimal maintenance and repair policies for railway lines, we use a previous design of risk model based on two factors: the criticality and the deterioration ratios of the facilities. Thus, our theory benefits from the Reliability Centered Management methodology application, but it also explicitly models uncertainty in characterizing a facility deterioration rate to decide the optimal policy to maintain the railway infrastructures. This may be the major contribution of this work. To verify the models presented, a computation study has been developed and tested for a real scenario: the railway line Villalba-Cercedilla in Madrid (Spain). Our results demonstrate again that applying any adaptive formulation, the cost of the railway lines maintenance shown is decreased. Moreover applying a Closed Loop Formulation the cost associated to the risk takes smaller values (40% less cost for the same risk than the deterministic approach), but with an Open Loop formulation the generated risk in the railway line is also smaller.

Ponencias de las Decimosextas Jornadas de Paralelismo celebradas del 13 al 16 de septiembre de 2005 en Granada El modelo de computación Grid ha evolucionado en los últimos años para proporcionar un entorno de computación de altas prestaciones en redes de área amplia. Sin embargo, uno de los mayores problemas se encuentra en las aplicaciones que hacen uso intensivo y masivo de datos. Como solución a los problemas de estas aplicaciones se ha utilizado la replicación. Sin embargo, la replicación clásica adolece de ciertos problemas como la adaptabilidad y la alta latencia del nuevo entorno. Por ello se propone un nuevo algoritmo de replicación y organización de datos que proporciona un acceso de altas prestaciones en un Data Grid. Publicado

Ponencias de las Decimocuartas Jornadas de Paralelismo de la Universidad Carlos III de Madrid celebradas el 15, 16 y 17 de septiembre de 2003 en Leganés, Madrid Este artículo presenta las técnicas de optimización utilizadas para acelerar la carga de definiciones de señales de un banco de pruebas de aviónica. La carga de datos en formato XML puede presentar problemas de rendimiento cuando se trata de grandes volumenes de datos. Bajo esas condiciones, es necesario buscar alternativas que permitan cargar los datos de forma más eficiente, sin imponer restricciones a los generadores de los datos utilizados como entrada. El artículo analiza posibles optimizaciones y valora las ventajas e inconvenientes de cada solución. Publicado

La E/S constituye en la actualidad uno de los principales cuellos de botella de los sistemas distribuidos de propósito general, debido al desequilibrio existente entre el tiempo de cómputo y de E/S. Una de las soluciones propuestas para este problema ha sido el uso de la E/S paralela. En esta área, se han originado un gran número de bibliotecas de E/S paralela y sistemas de ficheros paralelos. Este tipo de sistemas adolecen de algunos defectos y carencias. Muchos de ellos están concebidos para máquinas paralelas y no se integran adecuadamente en entornos distribuidos y clusters. El uso intensivo de clusters de estaciones de trabajo durante estos últimos años hace que este tipo de sistemas no sean adecuados en el escenario de computación actual. Otros sistemas, que se adaptan a este tipo de entornos, no incluyen capacidades de reconfiguración dinámica, por lo que tienen una funcionalidad limitada. Por último, la mayoría de los sistemas de E/S que utilizan diferentes optimizaciones de E/S, no ofrecen flexibilidad a las aplicaciones para hacer uso de las mismas, intentando ocultar al usuario este tipo de técnicas. No obstante, a fin de optimizar las operaciones de E/S, es importante que las aplicaciones sean capaces de describir sus patrones de acceso, interactuando con el sistema de E/S. En otro ámbito, dentro del área de los sistemas distribuidos se encuentra el paradigma de agentes, que permite dotar a las aplicaciones de un conjunto de propiedades muy adecuadas para su adaptación a entornos complejos y dinámicos. Las características de este paradigma lo hacen a priori prometedor para abordar algunos de los problemas existentes en el campo de la E/S paralela. Esta tesis propone una solución a la problemática actual de E/S a través de tres líneas principales: (i) el uso de la teoría de agentes en sistemas de E/S de alto rendimiento, (ii) la definición de un formalismo que permita la reconfiguración dinámica de nodos de almacenamiento en un cluster y (iii) el uso de técnicas de optimización de E/S configurables y orientadas a las aplicaciones.

Este proyecto esta centrado en mejorar la rentabilidad de inversionistas que realizan operaciones en los mercados bursátiles. Mediante una herramienta donde el inversor tenga a mano información actualizada sobre lo que está pasando en el mercado y en sus posiciones actuales. Aprovechando técnicas de aprendizaje automático se trata de aportar un mayor acierto en las operaciones para aumentar la rentabilidad anual. La estrategia de inversión utilizada esta basada en el largo plazo concretamente en la filosofía Value Investing, que se basa en realizar un análisis de los ratios financieros de las empresas o fundamentales de las compañías. En contraposición al Trading que se basa en el análisis técnico del precio y el volumen donde las operaciones se hacen en cortos periodos de tiempo (minutos, horas o días). El desarrollo culminará con una herramienta visual con un cuadro de mandos donde se mostrará las posiciones actuales del portfolio, rendimiento de la cartera, sugerencias de inversión basadas en predicciones por Machine Learning y un indicador de estado predictivo de compra, venta o mantener según el estado actual del mercado., Universidad Europea de Madrid, Máster Universitario en Análisis de Datos Masivos (Big Data), Virtual

The global ecosystem of information technology (IT) is in transition to a new generation of applications that require more and more intensive data acquisition, processing and storage systems. As a result of that change towards data intensive computing, there is a growing overlap between high performance computing (HPC) and Big Data techniques in applications, since many HPC applications produce large volumes of data, and Big Data needs HPC capabilities. The hypothesis of this PhD. thesis is that the potential interoperability and convergence of the HPC and Big Data systems are crucial for the future, being essential the unification of both paradigms to address a broad spectrum of research domains. For this reason, the main objective of this Phd. thesis is purposing and developing a monitoring system to allow the HPC and Big Data convergence, thanks to giving information about behaviors of applications in a system which execute both kind of them, giving information to improve scalability, data locality, and to allow adaptability to large scale computers. To achieve this goal, this work is focused on the design of resource monitoring and discovery to exploit parallelism at all levels. These collected data are disseminated to facilitate global improvements at the whole system, and, thus, avoid mismatches between layers. The result is a two-level monitoring framework (both at node and application level) with a low computational load, scalable, and that can communicate with different modules thanks to an API provided for this purpose. All data collected is disseminated to facilitate the implementation of improvements globally throughout the system, and thus avoid mismatches between layers, which combined with the techniques applied to deal with fault tolerance, makes the system robust and with high availability. On the other hand, the developed framework includes a task scheduler capable of managing the launch of applications, their migration between nodes, as well as the possibility of dynamically increasing or decreasing the number of processes. All these thanks to the cooperation with other modules that are integrated into LIMITLESS, and whose objective is to optimize the execution of a stack of applications based on multi-criteria policies. This scheduling mode is called coarse-grain scheduling based on monitoring. For better performance and in order to further reduce the overhead during the monitorization, different optimizations have been applied at different levels to try to reduce communications between components, while trying to avoid the loss of information. To achieve this objective, data filtering techniques, Machine Learning (ML) algorithms, and Neural Networks (NN) have been used. In order to improve the scheduling process and to design new multi-criteria scheduling policies, the monitoring information has been combined with other ML algorithms to identify (through classification algorithms) the applications and their execution phases, doing offline profiling. Thanks to this feature, LIMITLESS can detect which phase is executing an application and tries to share the computational resources with other applications that are compatible (there is no performance degradation between them when both are running at the same time). This feature is called fine-grain scheduling, and can reduce the makespan of the use cases while makes efficient use of the computational resources that other applications do not use. El ecosistema global de las tecnologías de la información (IT) se encuentra en transición a una nueva generación de aplicaciones que requieren sistemas de adquisición de datos, procesamiento y almacenamiento cada vez más intensivo. Como resultado de ese cambio hacia la computación intensiva de datos, existe una superposición, cada vez mayor, entre la computación de alto rendimiento (HPC) y las técnicas Big Data en las aplicaciones, pues muchas aplicaciones HPC producen grandes volúmenes de datos, y Big Data necesita capacidades HPC. La hipótesis de esta tesis es que hay un gran potencial en la interoperabilidad y convergencia de los sistemas HPC y Big Data, siendo crucial para el futuro tratar una unificación de ambos para hacer frente a un amplio espectro de problemas de investigación. Por lo tanto, el objetivo principal de esta tesis es la propuesta y desarrollo de un sistema de monitorización que facilite la convergencia de los paradigmas HPC y Big Data gracias a la provisión de datos sobre el comportamiento de las aplicaciones en un entorno en el que se pueden ejecutar aplicaciones de ambos mundos, ofreciendo información útil para mejorar la escalabilidad, la explotación de la localidad de datos y la adaptabilidad en los computadores de gran escala. Para lograr este objetivo, el foco se ha centrado en el diseño de mecanismos de monitorización y localización de recursos para explotar el paralelismo en todos los niveles de la pila del software. El resultado es un framework de monitorización en dos niveles (tanto a nivel de nodo como de aplicación) con una baja carga computacional, escalable, y que se puede comunicar con distintos módulos gracias a una API proporcionada para tal objetivo. Todos datos recolectados se difunden para facilitar la realización de mejoras de manera global en todo el sistema, y así evitar desajustes entre capas, lo que combinado con las técnicas aplicadas para lidiar con la tolerancia a fallos, hace que el sistema sea robusto y con una alta disponibilidad. Por otro lado, el framework desarrollado incluye un planificador de tareas capaz de gestionar el lanzamiento de aplicaciones, la migración de las mismas entre nodos, además de la posibilidad de incrementar o disminuir su número de procesos de forma dinámica. Todo ello gracias a la cooperación con otros módulos que se integran en LIMITLESS, y cuyo objetivo es optimizar la ejecución de una pila de aplicaciones en base a políticas multicriterio. Esta funcionalidad se llama planificación de grano grueso. Para un mejor desempeño y con el objetivo de reducir más aún la carga durante la ejecución, se han aplicado distintas optimizaciones en distintos niveles para tratar de reducir las comunicaciones entre componentes, a la vez que se trata de evitar la pérdida de información. Para lograr este objetivo se ha hecho uso de técnicas de filtrado de datos, algoritmos de Machine Learning (ML), y Redes Neuronales (NN). Finalmente, para obtener mejores resultados en la planificación de aplicaciones y para diseñar nuevas políticas de planificación multi-criterio, los datos de monitorización recolectados han sido combinados con nuevos algoritmos de ML para identificar (por medio de algoritmos de clasificación) aplicaciones y sus fases de ejecución. Todo ello realizando tareas de profiling offline. Gracias a estas técnicas, LIMITLESS puede detectar en qué fase de su ejecución se encuentra una determinada aplicación e intentar compartir los recursos de computacionales con otras aplicaciones que sean compatibles (no se produce una degradación del rendimiento entre ellas cuando ambas se ejecutan a la vez en el mismo nodo). Esta funcionalidad se llama planificación de grano fino y puede reducir el tiempo total de ejecución de la pila de aplicaciones en los casos de uso porque realiza un uso más eficiente de los recursos de las máquinas. This PhD dissertation has been partially supported by the Spanish Ministry of Science and Innovation under an FPI fellowship associated to a National Project with reference TIN2016-79637-P (from July 1, 2018 to October 10, 2021) Programa de Doctorado en Ciencia y Tecnología Informática por la Universidad Carlos III de Madrid Presidente: Félix García Carballeira.- Secretario: Pedro Ángel Cuenca Castillo.- Vocal: María Cristina V. Marinescu

El análisis de cantidades de datos cada vez más voluminosas, veloces y variadas, aunque sea a distintas velocidades, no es solo una realidad patente y permanente. También trasciende meras aplicaciones de índice matemático, estadístico y contable-financiero, aunque nada esté absolutamente disperso. La digitalización de la sociedad y la mayo dependencia de los individuos de la “red de redes”, tanto para bien como para mal nos permiten extraer información que permita elaborar un perfil de cierta envergadura sobre estos, sin necesidad de someterle a un complejo interrogatorio (por ejemplo). Esto mismo nos motiva a elaborar un prototipado que, de manera desinteresada y plenamente constructiva, contribuya al desarrollo de la telemedicina y otras aplicaciones tecnológicas relacionadas con las distintas especialidades de la misma (e-Health, obedeciendo al anglicismo). En nuestro caso, Psiquiatría y Psicología serán las más beneficiadas. Concretamente, implementamos un sistema de análisis de sentimientos mediante aprendizaje supervisado, a fin de clasificar por medio de la predicción, partiendo de un conjunto de datos entrenado. Queremos que se pueda clasificar, con la mayor exactitud posible, unos mensajes, según la positividad o la negatividad sentimental que quede reflejada en estos. Al mismo tiempo, se aplica una serie de técnicas de visualización que facilitan al interesado, en principio un facultativo sanitario, la comprensión de lo que no dejan de ser indicadores que le ayuden a determinar el diagnóstico de un posible paciente con problemas psiquiátricos, si es que, simplemente, no se trata de prevenir otros males mayores., Universidad Europea de Madrid, Máster Universitario en Análisis de Datos Masivos (Big Data), Virtual

Ossas (Operating Systems Scheduling Algorithms Simulator) son las siglas bajo las que se denomina a las múltiples herramientas que componen una solución completa para simular diferentes escenarios en cuanto a la planifcación de procesos en Sistemas Ope- rativos, mostrar el estado del sistema en cada momento y ejecutar el simulador en las plataformas más usadas en la actualidad. La pieza fundamental es la librería (ossas-lib) que contiene la lógica de la simulación, independiente de cualquier interfaz, permite su integración en cualquier entorno que per- mita la ejecución de código javascript, encontrado hoy en día en la web, plataformas de escritorio y plataformas móviles. Alrededor de esta librería se ha creado una aplicación Angular (ossas-ng), que propor- ciona una interfaz para poder interactuar con el simulador. Esta aplicación es altamente modular, ya que esta arquitectura es una parte integral del desarrollo con Angular, lo que permite su futura expansión de forma sencilla. Para llevar una aplicación web a las plataformas de escritorio se ha creado una apli- cación basada en Electron JS (ossas-electron). Esta forma de distribución añade una gran carga computacional debido a que se debe ejecutar una pequeña instancia de un navegador web para ejecutar la aplicación como si fuese nativa, sin embargo las ventajas obtenidas superan, en este caso, esta carga extra. Gracias a esta solución de herramientas se permite llevar un simulador de algoritmos de planifcación de sistemas operativos a cualquier dispositivo que pueda ejecutar un nave- gador web y permitir que cualquier persona pueda entender los conceptos de planifcación de procesos de una forma interactiva y visual. Ingeniería Informática

Mención Internacional en el título de doctor Convergence between high-performance computing (HPC) and Big Data analytics (BDA) is currently an established research area that spawned new opportunities for unifying the platformlayer and data abstractions in these ecosystems. This thesis builds on the hypothesis that HPC-BDA convergence at platform level can be attained by enabling runtime interoperability in a way that preserves BDA platform usability and productivity, exploits HPC scalability and performance, and expands both BDA and HPC capabilities to cope with prospect hybrid application models. The goal is to architect an abstract system that enables the interoperability of established BDA and HPC runtimes. In order to exploit the benefits of BDA data-centric paradigms, this thesis presents a data-centric transformation methodology to allow process-centric workloads the interaction with BDA platforms and storage infrastructures. Furthermore, an architecture to achieve full runtime interoperability is proposed. It reflects the key design features that interest both the HPC and BDA communities, and includes an abstract data collection and operational model that generates a unified interface for hybrid applications. It also incorporates a mechanism to transfer each stage of the application to the appropriate runtime. This architecture can be implemented in different ways depending on the process- and data-centric runtimes of choice, and the mechanisms put in place to effectively meet the requirements of the architecture. The Spark-DIY platformis introduced as a possible implementation. It preserves the interfaces and execution environment of the popular BDA platformApache Spark –thus making it compatible with any Spark-based application and tool– while providing efficient communication and kernel execution via DIY, a powerful communication pattern library built on top of MPI. Finally, these solutions are analysed in terms of performance by applying them to a representative use case, EnKF-HGS. This application is a clear example of how current HPC simulations are evolving towards hybrid HPC-BDA applications, integrating HPC simulations within a BDA environment. Other auxiliary use cases –like an application from the railway domain and a BDA benchmark operator– are also introduced to support other specific contributions of this thesis. La convergencia entre la computación de altas prestaciones (HPC) y el análisis de macrodatos (BDA) es actualmente un área de investigación establecida que ha generado nuevas oportunidades para la unificación de la capa de plataforma y las abstracciones de datos en estos ecosistemas. Esta tesis desarrolla la hipótesis de que la convergencia HPC-BDA a nivel de plataforma puede ser obtenida con la habilitación de mecanismos de interoperabilidad entre entornos de ejecución, de modo que se preserve la usabilidad y productividad de las plataformas BDA, se explote la escalabilidad y rendimiento de HPC, y se expandan las capacidades de HPC y BDA para tratar futuros modelos híbridos de aplicación. El objetivo es desarrollar un sistema abstracto que permita la interoperabilidad de entornos de ejecución ya establecidos en los ecosistemas BDA y HPC. Con el fin de explotar los beneficios de los paradigmas orientados a datos en BDA, esta tesis presenta una metodología de transformación también orientada a datos que permite a las aplicaciones orientadas a proceso interactuar con plataformas BDA y sus correspondientes infraestructuras de almacenamiento. Además, se propone una arquitectura para obtener interoperabilidad total entre entornos de ejecución. Ésta refleja las características de diseño clave que interesan a las comunidades BDA y HPC, e incluye una abstracción de colección de datos y modelo operacional que genera una interfaz unificada para aplicaciones híbridas. Además, incorpora un mecanismo para transferir cada etapa de la aplicación al entorno de ejecución adecuado. Esta arquitectura puede ser implementada de distintas maneras dependiendo de los entornos de ejecución orientados a datos y proceso seleccionados, y las tcnicas utilizadas para cumplir de manera efectiva con los requisitos de la arquitectura. La plataforma Spark-DIY se introduce como posible implementación. Preserva las interfaces y entorno de ejecución de la popular plataforma BDA Apache Spark –haciéndola compatible con cualquier aplicación o herramienta basada en Spark–, mientras provee comunicación y ejecución eficiente de núcleos de simulación y análisis a través de DIY, una potente biblioteca de patrones de comunicación construida sobre MPI. Finalmente, estas soluciones son analizadas en términos de rendimiento al aplicarlas a un caso de uso representativo, EnKF-HGS. Esta aplicación es un ejemplo claro de cómo las simulaciones HPC están evolucionando hacia aplicaciones HPC-BDA híbridas, integrando simulaciones HPC dentro de un entorno BDA. Otros casos de uso auxiliares –como una aplicación del ámbito ferroviario y un operador referente de BDA– son introducidos para apoyar otras contribuciones específicas de esta tesis. Programa de Doctorado en Ciencia y Tecnología Informática por la Universidad Carlos III de Madrid Presidente: Laurent Lefevre.- Secretario: David Expósito Singh.- Vocal: Mª de los Santos Pérez Hernández

Mención Internacional en el título de doctor Medical image processing is an interdisciplinary field in which multiple research areas are involved: image acquisition, scanner design, image reconstruction algorithms, visualization, etc. X-Ray Computed Tomography (CT) is a medical imaging modality based on the attenuation suffered by the X-rays as they pass through the body. Intrinsic differences in attenuation properties of bone, air, and soft tissue result in high-contrast images of anatomical structures. The main objective of CT is to obtain tomographic images from radiographs acquired using X-Ray scanners. The process of building a 3D image or volume from the 2D radiographs is known as reconstruction. One of the latest trends in CT is the reduction of the radiation dose delivered to patients through the decrease of the amount of acquired data. This reduction results in artefacts in the final images if conventional reconstruction methods are used, making it advisable to employ iterative reconstruction algorithms. There are numerous reconstruction algorithms available, from which we can highlight two specific types: traditional algorithms, which are fast but do not enable the obtaining of high quality images in situations of limited data; and iterative algorithms, slower but more reliable when traditional methods do not reach the quality standard requirements. One of the priorities of reconstruction is the obtaining of the final images in near real time, in order to reduce the time spent in diagnosis. To accomplish this objective, new high performance techniques and methods for accelerating these types of algorithms are needed. This thesis addresses the challenges of both traditional and iterative reconstruction algorithms, regarding acceleration and image quality. One common approach for accelerating these algorithms is the usage of shared-memory and heterogeneous architectures. In this thesis, we propose a novel simulation/reconstruction framework, namely FUX-Sim. This framework follows the hypothesis that the development of new flexible X-ray systems can benefit from computer simulations, which may also enable performance to be checked before expensive real systems are implemented. Its modular design abstracts the complexities of programming for accelerated devices to facilitate the development and evaluation of the different configurations and geometries available. In order to obtain near real execution times, low-level optimizations for the main components of the framework are provided for Graphics Processing Unit (GPU) architectures. Other alternative tackled in this thesis is the acceleration of iterative reconstruction algorithms by using distributed memory architectures. We present a novel architecture that unifies the two most important computing paradigms for scientific computing nowadays: High Performance Computing (HPC). The proposed architecture combines Big Data frameworks with the advantages of accelerated computing. The proposed methods presented in this thesis provide more flexible scanner configurations as they offer an accelerated solution. Regarding performance, our approach is as competitive as the solutions found in the literature. Additionally, we demonstrate that our solution scales with the size of the problem, enabling the reconstruction of high resolution images. El procesamiento de imágenes médicas es un campo interdisciplinario en el que participan múltiples áreas de investigación como la adquisición de imágenes, diseño de escáneres, algoritmos de reconstrucción de imágenes, visualización, etc. La tomografía computarizada (TC) de rayos X es una modalidad de imágen médica basada en el cálculo de la atenuación sufrida por los rayos X a medida que pasan por el cuerpo a escanear. Las diferencias intrínsecas en la atenuación de hueso, aire y tejido blando dan como resultado imágenes de alto contraste de estas estructuras anatómicas. El objetivo principal de la TC es obtener imágenes tomográficas a partir estas radiografías obtenidas mediante escáneres de rayos X. El proceso de construir una imagen o volumen en 3D a partir de las radiografías 2D se conoce como reconstrucción. Una de las últimas tendencias en la tomografía computarizada es la reducción de la dosis de radiación administrada a los pacientes a través de la reducción de la cantidad de datos adquiridos. Esta reducción da como resultado artefactos en las imágenes finales si se utilizan métodos de reconstrucción convencionales, por lo que es aconsejable emplear algoritmos de reconstrucción iterativos. Existen numerosos algoritmos de reconstrucción disponibles a partir de los cuales podemos destacar dos categorías: algoritmos tradicionales, rápidos pero no permiten obtener imágenes de alta calidad en situaciones en las que los datos son limitados; y algoritmos iterativos, más lentos pero más estables en situaciones donde los métodos tradicionales no alcanzan los requisitos en cuanto a la calidad de la imagen. Una de las prioridades de la reconstrucción es la obtención de las imágenes finales en tiempo casi real, con el fin de reducir el tiempo de diagnóstico. Para lograr este objetivo, se necesitan nuevas técnicas y métodos de alto rendimiento para acelerar estos algoritmos. Esta tesis aborda los desafíos de los algoritmos de reconstrucción tradicionales e iterativos, con respecto a la aceleración y la calidad de imagen. Un enfoque común para acelerar estos algoritmos es el uso de arquitecturas de memoria compartida y heterogéneas. En esta tesis, proponemos un nuevo sistema de simulación/reconstrucción, llamado FUX-Sim. Este sistema se construye alrededor de la hipótesis de que el desarrollo de nuevos sistemas de rayos X flexibles puede beneficiarse de las simulaciones por computador, en los que también se puede realizar un control del rendimiento de los nuevos sistemas a desarrollar antes de su implementación física. Su diseño modular abstrae las complejidades de la programación para aceleradores con el objetivo de facilitar el desarrollo y la evaluación de las diferentes configuraciones y geometrías disponibles. Para obtener ejecuciones en casi tiempo real, se proporcionan optimizaciones de bajo nivel para los componentes principales del sistema en las arquitecturas GPU. Otra alternativa abordada en esta tesis es la aceleración de los algoritmos de reconstrucción iterativa mediante el uso de arquitecturas de memoria distribuidas. Presentamos una arquitectura novedosa que unifica los dos paradigmas informáticos más importantes en la actualidad: computación de alto rendimiento (HPC) y Big Data. La arquitectura propuesta combina sistemas Big Data con las ventajas de los dispositivos aceleradores. Los métodos propuestos presentados en esta tesis proporcionan configuraciones de escáner más flexibles y ofrecen una solución acelerada. En cuanto al rendimiento, nuestro enfoque es tan competitivo como las soluciones encontradas en la literatura. Además, demostramos que nuestra solución escala con el tamaño del problema, lo que permite la reconstrucción de imágenes de alta resolución. This work has been mainly funded thanks to a FPU fellowship (FPU14/03875) from the Spanish Ministry of Education. It has also been partially supported by other grants: • DPI2016-79075-R. “Nuevos escenarios de tomografía por rayos X”, from the Spanish Ministry of Economy and Competitiveness. • TIN2016-79637-P Towards unification of HPC and Big Data Paradigms from the Spanish Ministry of Economy and Competitiveness. • Short-term scientific missions (STSM) grant from NESUS COST Action IC1305. • TIN2013-41350-P, Scalable Data Management Techniques for High-End Computing Systems from the Spanish Ministry of Economy and Competitiveness. • RTC-2014-3028-1 NECRA Nuevos escenarios clinicos con radiología avanzada from the Spanish Ministry of Economy and Competitiveness. Programa Oficial de Doctorado en Ciencia y Tecnología Informática Presidente: José Daniel García Sánchez.- Secretario: Katzlin Olcoz Herrero.- Vocal: Domenico Talia

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. As we move towards ultrascale computing, computer architecture is bound to see dramatic changes. Multiple nodes, with or without shared memory, multicore and accelerators (GPUs, FPGAs) will be the norm. For many problems, such as finite difference numerical simulations, the array used to represent a perfect match between the user level code and the hardware architecture’s uniform memory access. Arrays, and to some extent multiarrays, are well supported by most programming languages. A standard compiler maps the array for uniform memory. Some programming models, such as partitioned global address space, allows mapping an array across distributed, yet for each partition, uniform memory. For ultrascale architectures, the simple mapping between user level (multi)array and distributed, non-uniform memory, will disappear. Here we propose an API for arrays, empowering the software developer to implement their own array-memory layout. Application code written towards the API will be independent of underlying architecture changes, thus easily ported to new architectures as they evolve. This research has in part been financed by The Research Council of Norway through the project Design of a Mouldable Programming Language (DMPL), and has received support from EU under the COST Program Action IC1305, Network for Sustainable Ultrascale Computing (NESUS).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The modern high-performance computing systems (HPCS) are composed of hundreds of thousand computational nodes. An effective resource allocation in HPCS is a subject for many scientific research investigations. Many programming models for effective resources allocation have been proposed. The main purpose of those models is to increase the parallel performance of the HPCS. This paper investigates the efficiency of parallel algorithm for resource management optimization based on Artificial Bee Colony (ABC) metaheuristic while solving a package of NP-complete problems on multi-processor platform.In order to achieve minimal parallelization overhead in each cluster node, a multi-level hybrid programming model is proposed that combines coarse-grain and fine-grain parallelism. Coarse-grain parallelism is achieved through domain decomposition by message passing among computational nodes using Message Passing Interface (MPI) and fine-grain parallelism is obtained by loop-level parallelism inside each computation node by compiler-based thread parallelization via Intel TBB. Parallel communications profiling is made and parallel performance parameters are evaluated on the basis of experimental results.

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The main goal of this paper is the suitability assessment of the OpenMP Accelerator Model (OMPAM) for porting a real-life scientific application to heterogeneous platforms containing a single Intel Xeon Phi coprocessor. This OpenMP extension is supported from version 4.0 of the standard, offering an unified directive-based programming model dedicated for massively parallel accelerators. In our study, we focus on applying the OMPAM extension together with the OpenMP tasks for a parallel application which implements the numerical model of alloy solidification. To map the application efficiently on target hybrid platforms using such constructs as omp target, omp target data and omp target update, we propose a decomposition of main tasks belonging to the computational core of the studied application. In consequence, the coprocessor is used to execute the major parallel workloads, while CPUs are responsible for executing a part of the application that do not require massively parallel resources. Effective overlapping computations with data transfers is another goal achieved in this way. The proposed approach allows us to execute the whole application 3.5 times faster than the original parallel version running on two CPUs. This research was conducted with the support of COST Action IC1305 (NESUS), as well as the National Science Centre (Poland) under grant no. UMO-2011/03/B/ST6/03500. The authors are grateful to the Czestochowa University of Technology for granting access to Intel Xeon Phi coprocessors provided by the MICLAB project no. POIG.02.03.00.24-093/13 (http://miclab.pl).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. The power consumption of data centers is becoming a crucial challenge in the context of the steadily increasing demand for computation. In this regard finding a way to improve energy efficiency of running applications in data centers is becoming a crucial trend. One method to improve the processor utilization is the consolidation of applications on physical servers. It is possible to run multiple jobs in parallel on the same machine, especially when their requirements regarding computation are smaller than the maximum processor performance. It reduces the number of servers in the data center required to handle multiple requests and therefore leads to energy usage reductions. In this paper, we introduce a realistic model of applications with deadlines executed in parallel on a server and competing for the shared resources and present an energy-aware algorithm which may be used to minimize the overall energy consumption of the servers. European Community's Seventh Framework Program This work is partially supported by EU under the COST Program Action 1305: Network for Sustainable Ultrascale Computing (NESUS). The research presented in this paper is partially funded by a grant from Polish National Science Center under award number 2013/08/A/ST6/00296. This research was supported by the EU Seventh Framework Programme FP7/2007–2013 under grant agreement no. FP7-ICT-2013-10 (609757).

Proceedings of: Third International Workshop on Sustainable Ultrascale Computing Systems (NESUS 2016). Sofia (Bulgaria), October, 6-7, 2016. In this paper we address the issue of managing different energy sources which supply green powered datacenters. The sources are scheduled based on a priority scheme, aiming to maximize the renewable energy utilization, minimize the energy used from the grid and optimize battery usage. Dynamic power capping technique is used to put a threshold on the drawn energy from the grid. The algorithm is implemented and tested in CloudSim simulator. Renewable energy is considered as solar energy. A workload scheduling algorithm is already implemented for higher renewable energy utilization. The results show that the proposed scheme is efficient and it is a promising direction in the field of the optimization in datacenters using renewable energy. The work presented in this paper has been supported by EU under the COST programme Action IC1305: Network for Sustainable Ultrascale Computing (NESUS).

In this work we will focus on overhead air switches design problem. The design of railway infrastructures is an important problem in the railway world, non-optimal designs cause limitations in the train speed and, most important, malfunctions and breakages. Most railway companies have regulations for the design of these elements. Those regulations have been defined by the experience, but, as far as we know, there are no computerized software tools that assist with the task of designing and testing optimal solutions for overhead switches. The aim of this thesis is the design, implementation, and evaluation of a simulator that that facilitates the exploration of all possible solutions space, looking for the set of optimal solutions in the shortest time and at the lowest possible cost. Simulators are frequently used in the world of rail infrastructure. Many of them only focus on simulated scenarios predefined by the users, analyzing the feasibility or otherwise of the proposed design. Throughout this thesis, we will propose a framework to design a complete simulator that be able to propose, simulate and evaluate multiple solutions. This framework is based on four pillars: compromise between simulation accuracy and complexity, automatic generation of possible solutions (automatic exploration of the solution space), consideration of all the actors involved in the design process (standards, additional restrictions, etc.), and finally, the expert’s knowledge and integration of optimization metrics. Once we defined the framework different deployment proposes are presented, one to be run in a single node, and one in a distributed system. In the first paradigm, one thread per CPU available in the system is launched. All the simulators are designed around this paradigm of parallelism. The second simulation approach will be designed to be deploy in a cluster with several nodes, MPI will be used for that purpose. Finally, after the implementation of each of the approaches, we will proceed to evaluate the performance of each of them, carrying out a comparison of time and cost. Two examples of real scenarios will be used. El diseño de agujas aéreas es un problema bastante complejo y critico dentro del proceso de diseño de sistemas ferroviarios. Un diseño no óptimo puede provocar limitaciones en el servicio, como menor velocidad de tránsito, y lo que es más importante, puede ser la causa principal de accidentes y averías. La mayoría de las compañías ferroviarias disponen de regulaciones para el diseño correcto de estas agujas aéreas. Todas estas regulaciones han sido definidas bajo décadas de experiencia, pero hasta donde sé, no existen aplicaciones software que ayuden en la tarea de diseñar y probar soluciones óptimas. Es en este punto donde se centra el objetivo de la tesis, el diseño, implementación y evaluación de un simulador capaz de explorar todo el posible espacio de soluciones buscando el conjunto de soluciones óptimas en el menor tiempo y con el menor coste posible. Los simuladores son utilizados frecuentemente en el mundo de la infraestructura ferroviaria. Muchos de ellos solo se centran en la simulación de escenarios preestablecidos por el usuario, analizando la viabilidad o no del diseño propuesto. A lo largo de esta tesis, se propondrá un framework que permita al simulador final ser capaz de proponer, simular y evaluar múltiples soluciones. El framework se basa en 4 pilares fundamentales, compromiso entre precisión en la simulación y la complejidad del simulador; generación automática de posibles soluciones (exploración automática del espacio de soluciones), consideración de todos los agentes que intervienen en el proceso de diseño (normativa, restricciones adicionales, etc.) y por último, el conocimiento del experto y la integración de métricas de optimización. Una vez definido el framework se presentaran varias opciones de implementación del simulador, en la primera de ellas se diseñará e implementara una versión con hilos pura. Se lanzara un hilo por cada CPU disponible en el sistema. Todo el simulador se diseñará en torno a este paradigma de paralelismo. En un segundo simulador, se aplicará un paradigma mucho más pensado para su despliegue en un cluster y no en un único nodo (como el paradigma inicial), para ello se empleara MPI. Con esta versión se podrá adaptar el simulador al cluster en el que se va a ejecutar. Por último, se va a emplear un paradigma basado en cloud computing. Para ello, según las necesidades del escenario a simular, se emplearán más o menos máquinas virtuales. Finalmente, tras la implementación de cada uno de los simuladores, se procederá a evaluar el rendimiento de cada uno de ellos, realizando para ello una comparativa de tiempo y coste. Se empleara para ello dos ejemplos de escenarios reales. Programa Oficial de Doctorado en Ciencia y Tecnología Informática Presidente: José Daniel García Sánchez.- Secretario: Antonio García Dopico.- Vocal: Juan Carlos Díaz Martín

Mención Internacional en el título de doctor, Many scientific areas make extensive use of computer simulations to study complex real-world processes. These computations are typically very resource-intensive and present scalability issues as experiments get larger, even in dedicated supercomputers since they are limited by their own hardware resources. Cloud computing raises as an option to move forward into the ideal unlimited scalability by providing virtually infinite resources, yet applications must be adapted to this paradigm. The major goal of this thesis is to analyze the suitability of performing simulations in clouds by performing a paradigm shift, from classic parallel approaches to data-centric models, in those applications where that is possible. The aim is to maintain the scalability achieved in traditional HPC infrastructures, while taking advantage of Cloud Computing paradigm features. The thesis also explores the characteristics that make simulators suitable or unsuitable to be deployed on HPC or Cloud infrastructures, defining a generic architecture and extracting common elements present among the majority of simulators. As result, we propose a generalist cloudification methodology based on the MapReduce paradigm to migrate high performance simulations into the cloud to provide greater scalability. We analysed its viability by applying it to a real engineering simulator and running the resulting implementation on HPC and cloud environments. Our evaluations will aim to show that the cloudified application is highly scalable and there is still a large margin to improve the theoretical model and its implementations, and also to extend it to a wider range of simulations., Muchas áreas de investigación hacen uso extensivo de simulaciones informáticas para estudiar procesos complejos del mundo real. Estas simulaciones suelen hacer uso intensivo de recursos, y presentan problemas de escalabilidad conforme los experimentos aumentan en tamaño incluso en clústeres, ya que estos están limitados por sus propios recursos hardware. Cloud Computing (computación en la nube) surge como alternativa para avanzar hacia el ideal de escalabilidad ilimitada mediante el aprovisionamiento de infinitos recursos (de forma virtual). No obstante, las aplicaciones deben ser adaptadas a este nuevo paradigma. La principal meta de esta tesis es analizar la idoneidad de realizar simulaciones en la nube mediante un cambio de paradigma, de las clásicas aproximaciones paralelas a nuevos modelos centrados en los datos, en aquellas aplicaciones donde esto sea posible. El objetivo es mantener la escalabilidad alcanzada en las tradicionales infraestructuras HPC, mientras se explotan las ventajas del paradigma de computación en la nube. La tesis explora las características que hacen a los simuladores ser o no adecuados para ser desplegados en infraestructuras clúster o en la nube, definiendo una arquitectura genérica y extrayendo elementos comunes presentes en la mayoría de los simuladores. Como resultado, proponemos una metodología genérica de cloudificación, basada en el paradigma MapReduce, para migrar simulaciones de alto rendimiento a la nube con el fin de proveer mayor escalabilidad. Analizamos su viabilidad aplicándola a un simulador real de ingeniería, y ejecutando la implementación resultante en entornos clúster y en la nube. Nuestras evaluaciones pretenden mostrar que la aplicación cloudificada es altamente escalable, y que existe un amplio margen para mejorar el modelo teórico y sus implementaciones, y para extenderlo a un rango más amplio de simulaciones., Administrador de Infraestructuras Ferroviarias (ADIF), Estudio y realización de programas de cálculo de pórticos rígidos de catenaria (CALPOR) y de sistema de simulación de montaje de agujas aéreas de línea aérea de contacto (SIA), JM/RS 3.6/4100.0685-9/00100 – Administrador de Infraestructuras Ferroviarias (ADIF), Proyecto para la Investigación sobre la aplicación de las TIC a la innovación de las diferentes infraestructuras correspondientes a las instalaciones de electrificación y suministro de energía (SIRTE), JM/RS 3.9/1500.0009/0-00000 – Spanish Ministry of Education, TIN2010-16497, Scalable Input/Output techniques for high-performance distributed and parallel computing environments – Spanish Ministry of Economics and Competitiveness, TIN2013-41350-P, Técnicas de gestión escalable de datos para high-end computing systems – European Union, COST Action IC1305, ”Network for Sustainable Ultrascale Computing Platforms” (NESUS) – European Union, COST Action IC0805, ”Open European Network for High Performance Computing on Complex Environments” – Spanish Ministry of Economics and Competitiveness, TIN2011-15734-E, Red de Computación de Altas Prestaciones sobre Arquitecturas Paralelas Heterogéneas (CAPAP-H), Programa Oficial de Doctorado en Ciencia y Tecnología Informática, Presidente: Domenica Talia.- Presidente: José Daniel García Sánchez.- Secretario: José Manuel Moya Fernández

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. Parallelism has become one of the most extended paradigms used to improve performance. Legacy source code needs to be re-written so that it can take advantage of multi-core and many-core computing devices, such as GPGPU, FPGA, DSP or specific accelerators. However, it forces software developers to adapt applications and coding mechanisms in order to exploit the available computing devices. It is a time consuming and error prone task that usually results in expensive and sub-optimal parallel software. In this work, we describe a parallel programming model, a set of annotating techniques and a static scheduling algorithm for parallel applications. Their purpose is to simplify the task of transforming sequential legacy code into parallel code capable of making full use of several different computing devices with the objetive of increasing performance, lowering energy consumption and increase the productivity of the developer. European Cooperation in Science and Technology. COST The work presented in this paper has been partially supported by EU under the COST programme Action IC1305, ’Network for Sustainable Ultrascale Computing (NESUS)’ The research leading to these results has received funding from the European Union Seventh Framework Programme (FP7/2007-2013) under grant agreement n. 609666 and by the Spanish Ministry of Economics and Competitiveness under the grant TIN2013-41350-P.

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. GPUs benefit from much more computation power with the same order of energy consumption than CPUs. Thanks to their massive data parallel architecture, GPUs can outperform CPUs, especially on Single Program Multiple Data (SPMD) programming paradigm on a large amount of data. Database engines are now everywhere, from different sizes and complexities, for multiple usages, embedded or distributed; in 2012, 500 million of SQLite active instances were estimated over the world. Our goal is to exploit the computation power of GPUs to improve performance of SQLite, which is a key software component of many applications and systems. In this paper, we introduce CuDB, a GPU-boosted in-memory database engine (IMDB) based on SQLite. The SQLite API remains unchanged, allowing developers to easily upgrade database engine from SQlite to CuDB even on already existing applications. Preliminary results show significant speedups of 70x with join queries on datasets of 1 million records. We also demonstrate the "memory bounded" character of GPU-databases and show the energy efficiency of our approach. European Cooperation in Science and Technology. COST

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. This paper describes the model designed for the instruction mapping tool, which can be used for generating the low level assembly code for the digital signal processing algorithms. The model is based on the Very Long Instruction Word architecture. The Texas Instrument TMS320C6678 was the pattern and finally was described with the created model. The paper is showing the parameters of the hardware resources and also the instruction set. European Cooperation in Science and Technology. COST

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. Cloud computing offers a wide range of resources and services through the Internet that can been used for various purposes. The rapid growth of cloud computing has exempted many companies and institutions from the burden of maintaining expensive hardware and software infrastructure. With characteristics like high scalability, availability and fault tolerance, cloud computing meet the new era needs for massive data processing at an affordable cost. In our doctoral research we intend to study, analyze, evaluate and make proposals in order to further improve the performance of cloud computing. European Cooperation in Science and Technology. COST

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. Nowadays Tables and Smart phones are equipped with low power processor. Some of them, like the NVIDIA Tegra SoC, also come with a GPU integrated so that both, the CPU and the GPU have access directly to the same RAM memory. In another vein, one the main limitations of microphone array algorithms for audio processing is the high computational cost required to reproduce real acoustics environments when real-time signal processing is absolutely required. One of these algorithms is the Beamforming Algorithm, which is used to recover acoustic signals from their observations when they are corrupted by noise, reverberation and other interfering signals. In order to achieve real-time processing executing this algorithm we have employed high performance libraries such as OPENBLAS, LAPACK, CUBLAS, PLASMA and MAGMA, and a particular tune programming for these mobile devices. European Cooperation in Science and Technology. COST

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. We present the work in progress for the PhD thesis titled “Enhancing the programmability and energy efficiency of storage in HPC and cloud environments”. In this thesis, we focus on studying and optimizing data movement across different layers of the operating system’s I/O stack. We study the power consumption during I/O-intensive workloads using sophisticated software and hardware instrumentation, collecting time series data from internal ATX power lines that feed every system component, and several run-time operating system metrics. Data exploration and data analysis reveal for each I/O access pattern various power and performance regimes. These regimes show how power is used by the system as data moved through the I/O stack. We use this knowledge to build I/O power models that are able to predict power consumption for different I/O workloads, and optimize the CPU device driver that manage performance states to obtain great power savings (over 30%). Finally, we develop new mechanisms and abstractions that allow co-located virtual machines to share data with each other more efficiently. Our virtualized data sharing solution reduces data movement among virtual domains, leading to energy savings I/O performance improvements. European Cooperation in Science and Technology. COST

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. Nowadays, images and videos have been present everywhere, they can come directly from camera, mobile devices or from other peoples that share their images and videos. The latter are used to illustrate different objects in a large number of situations. This makes from image and video processing algorithms a very important tool used for various domains related to computer vision such as video surveillance, medical imaging and database (images and videos) indexation methods. The performance of these algorithms have been so reduced due the the high intensive computation required when using new image and video standards. In this paper, we propose a new framework that allows users to select in a smart and efficient way the processing units (GPU or/and CPU) within heterogeneous systems, when treating different kinds of multimedia objects : single image, multiple images, multiple videos and video in real time. The framework disposes of different image and video primitive functions that are implemented on GPU, such as shape (silhouette) detection, motion tracking using optical flow estimation, edges and corners detection. We have exploited these functions for several situations such as indexing videos, segmenting vertebrae in in X-ray and MR images, detecting and localizing event in multi-user scenarios. Experimentation showed interesting accelerations ranging from 6 to 118, by comparison with sequential implementations. Moreover, the parallel and heterogeneous implementations offered lower power consumption as a result for the fast treatment. European Cooperation in Science and Technology. COST

Proceedings of the First PhD Symposium on Sustainable Ultrascale Computing Systems (NESUS PhD 2016) Timisoara, Romania. February 8-11, 2016. Current GPUs (Graphic Processing Units) can obtain high computational performance in scientific applications. Nevertheless, programmers have to use suitable parallel algorithms for these architectures and have to consider optimization techniques in the implementation in order to achieve that performance. This thesis is focused on designing and implementing parallel prefix algorithms into GPU architectures with little effort. For that, we have developed a very optimized library called BPLG (Tuning Butterfly Processing Library for GPUs) and based on a set of building blocks that enable to easily design well-known algorithms such as FFT, tridiagonal systems solvers, scan operator, sorting or signal processing. This library is designed under a tuning methodology based on two-stages indentified as GPU resource analysis and operator string manipulation. Specifically, this strategy is focused on a set of parallel prefix algorithms that can be represented according to a set of common permutations of the digits of each of its element indices [4], denoted as Index-Digit (ID) algorithms. So far, the proposed methodology has obtained very good results with respect to state-of-art libraries, as CUFFT, CUSPARSE, CUDPP or ModernGPU. European Cooperation in Science and Technology. COST