Distributed processing of large remote sensing images using MapReduce - A case of Edge Detection

Dissertation submitted in partial fulfillment of the requirements for the Degree of Master of Science in Geospatial Technologies.

A Framework for Building Hypercubes Using MapReduce

The European Space Agency's Gaia mission will create the largest and most precise three dimensional chart of our galaxy (the Milky Way), by providing unprecedented position, parallax, proper motion, and radial velocity measurements for about one billion stars. The resulting catalogue will be made available to the scientific community and will be analyzed in many different ways, including the production of a variety of statistics. The latter will often entail the generation of multidimensional histograms and hypercubes as part of the precomputed statistics for each data release, or for scientific analysis involving either the final data products or the raw data coming from the satellite instruments. In this paper we present and analyze a generic framework that allows the hypercube generation to be easily done within a MapReduce infrastructure, providing all the advantages of the new Big Data analysis paradigmbut without dealing with any specific interface to the lower level distributed system implementation (Hadoop). Furthermore, we show how executing the framework for different data storage model configurations (i.e. row or column oriented) and compression techniques can considerably improve the response time of this type of workload for the currently available simulated data of the mission. In addition, we put forward the advantages and shortcomings of the deployment of the framework on a public cloud provider, benchmark against other popular solutions available (that are not always the best for such ad-hoc applications), and describe some user experiences with the framework, which was employed for a number of dedicated astronomical data analysis techniques workshops.

A Framework for Building Hypercubes Using MapReduce

The European Space Agency's Gaia mission will create the largest and most precise three dimensional chart of our galaxy (the Milky Way), by providing unprecedented position, parallax, proper motion, and radial velocity measurements for about one billion stars. The resulting catalogue will be made available to the scientific community and will be analyzed in many different ways, including the production of a variety of statistics. The latter will often entail the generation of multidimensional histograms and hypercubes as part of the precomputed statistics for each data release, or for scientific analysis involving either the final data products or the raw data coming from the satellite instruments. In this paper we present and analyze a generic framework that allows the hypercube generation to be easily done within a MapReduce infrastructure, providing all the advantages of the new Big Data analysis paradigmbut without dealing with any specific interface to the lower level distributed system implementation (Hadoop). Furthermore, we show how executing the framework for different data storage model configurations (i.e. row or column oriented) and compression techniques can considerably improve the response time of this type of workload for the currently available simulated data of the mission. In addition, we put forward the advantages and shortcomings of the deployment of the framework on a public cloud provider, benchmark against other popular solutions available (that are not always the best for such ad-hoc applications), and describe some user experiences with the framework, which was employed for a number of dedicated astronomical data analysis techniques workshops.

A scalable expressive ensemble learning using Random Prism: a MapReduce approach

The induction of classification rules from previously unseen examples is one of the most important data mining tasks in science as well as commercial applications. In order to reduce the influence of noise in the data, ensemble learners are often applied. However, most ensemble learners are based on decision tree classifiers which are affected by noise. The Random Prism classifier has recently been proposed as an alternative to the popular Random Forests classifier, which is based on decision trees. Random Prism is based on the Prism family of algorithms, which is more robust to noise. However, like most ensemble classification approaches, Random Prism also does not scale well on large training data. This paper presents a thorough discussion of Random Prism and a recently proposed parallel version of it called Parallel Random Prism. Parallel Random Prism is based on the MapReduce programming paradigm. The paper provides, for the first time, novel theoretical analysis of the proposed technique and in-depth experimental study that show that Parallel Random Prism scales well on a large number of training examples, a large number of data features and a large number of processors. Expressiveness of decision rules that our technique produces makes it a natural choice for Big Data applications where informed decision making increases the user’s trust in the system.

Enhanced Failure Detection Mechanism in MapReduce

The popularity of MapReduce programming model has increased interest in the research community for its improvement. Among the other directions, the point of fault tolerance, concretely the failure detection issue seems to be a crucial one, but that until now has not reached its satisfying level. Motivated by this, I decided to devote my main research during this period into having a prototype system architecture of MapReduce framework with a new failure detection service, containing both analytical (theoretical) and implementation part. I am confident that this work should lead the way for further contributions in detecting failures to any NoSQL App frameworks, and cloud storage systems in general.

Optimizing the reliability and resource efficiency of MapReduce-based systems

Debido al gran incremento de datos digitales que ha tenido lugar en los últimos años, ha surgido un nuevo paradigma de computación paralela para el procesamiento eficiente de grandes volúmenes de datos. Muchos de los sistemas basados en este paradigma, también llamados sistemas de computación intensiva de datos, siguen el modelo de programación de Google MapReduce. La principal ventaja de los sistemas MapReduce es que se basan en la idea de enviar la computación donde residen los datos, tratando de proporcionar escalabilidad y eficiencia. En escenarios libres de fallo, estos sistemas generalmente logran buenos resultados. Sin embargo, la mayoría de escenarios donde se utilizan, se caracterizan por la existencia de fallos. Por tanto, estas plataformas suelen incorporar características de tolerancia a fallos y fiabilidad. Por otro lado, es reconocido que las mejoras en confiabilidad vienen asociadas a costes adicionales en recursos. Esto es razonable y los proveedores que ofrecen este tipo de infraestructuras son conscientes de ello. No obstante, no todos los enfoques proporcionan la misma solución de compromiso entre las capacidades de tolerancia a fallo (o de manera general, las capacidades de fiabilidad) y su coste. Esta tesis ha tratado la problemática de la coexistencia entre fiabilidad y eficiencia de los recursos en los sistemas basados en el paradigma MapReduce, a través de metodologías que introducen el mínimo coste, garantizando un nivel adecuado de fiabilidad. Para lograr esto, se ha propuesto: (i) la formalización de una abstracción de detección de fallos; (ii) una solución alternativa a los puntos únicos de fallo de estas plataformas, y, finalmente, (iii) un nuevo sistema de asignación de recursos basado en retroalimentación a nivel de contenedores. Estas contribuciones genéricas han sido evaluadas tomando como referencia la arquitectura Hadoop YARN, que, hoy en día, es la plataforma de referencia en la comunidad de los sistemas de computación intensiva de datos. En la tesis se demuestra cómo todas las contribuciones de la misma superan a Hadoop YARN tanto en fiabilidad como en eficiencia de los recursos utilizados. ABSTRACT Due to the increase of huge data volumes, a new parallel computing paradigm to process big data in an efficient way has arisen. Many of these systems, called dataintensive computing systems, follow the Google MapReduce programming model. The main advantage of these systems is based on the idea of sending the computation where the data resides, trying to provide scalability and efficiency. In failure-free scenarios, these frameworks usually achieve good results. However, these ones are not realistic scenarios. Consequently, these frameworks exhibit some fault tolerance and dependability techniques as built-in features. On the other hand, dependability improvements are known to imply additional resource costs. This is reasonable and providers offering these infrastructures are aware of this. Nevertheless, not all the approaches provide the same tradeoff between fault tolerant capabilities (or more generally, reliability capabilities) and cost. In this thesis, we have addressed the coexistence between reliability and resource efficiency in MapReduce-based systems, looking for methodologies that introduce the minimal cost and guarantee an appropriate level of reliability. In order to achieve this, we have proposed: (i) a formalization of a failure detector abstraction; (ii) an alternative solution to single points of failure of these frameworks, and finally (iii) a novel feedback-based resource allocation system at the container level. Finally, our generic contributions have been instantiated for the Hadoop YARN architecture, which is the state-of-the-art framework in the data-intensive computing systems community nowadays. The thesis demonstrates how all our approaches outperform Hadoop YARN in terms of reliability and resource efficiency.

Large scale data processing using MapReduce

As massive data sets become increasingly available, people are facing the problem of how to effectively process and understand these data. Traditional sequential computing models are giving way to parallel and distributed computing models, such as MapReduce, both due to the large size of the data sets and their high dimensionality. This dissertation, as in the same direction of other researches that are based on MapReduce, tries to develop effective techniques and applications using MapReduce that can help people solve large-scale problems. Three different problems are tackled in the dissertation. The first one deals with processing terabytes of raster data in a spatial data management system. Aerial imagery files are broken into tiles to enable data parallel computation. The second and third problems deal with dimension reduction techniques that can be used to handle data sets of high dimensionality. Three variants of the nonnegative matrix factorization technique are scaled up to factorize matrices of dimensions in the order of millions in MapReduce based on different matrix multiplication implementations. Two algorithms, which compute CANDECOMP/PARAFAC and Tucker tensor decompositions respectively, are parallelized in MapReduce based on carefully partitioning the data and arranging the computation to maximize data locality and parallelism.

sMapReduce: a programming pattern for wireless sensor networks

Wireless Sensor Networks (WSNs) are increasingly used in various application domains like home-automation, agriculture, industries and infrastructure monitoring. As applications tend to leverage larger geographical deployments of sensor networks, the availability of an intuitive and user friendly programming abstraction becomes a crucial factor in enabling faster and more efficient development, and reprogramming of applications. We propose a programming pattern named sMapReduce, inspired by the Google MapReduce framework, for mapping application behaviors on to a sensor network and enabling complex data aggregation. The proposed pattern requires a user to create a network-level application in two functions: sMap and Reduce, in order to abstract away from the low-level details without sacrificing the control to develop complex logic. Such a two-fold division of programming logic is a natural-fit to typical sensor networking operation which makes sensing and topological modalities accessible to the user.

Data analytics in the cloud with flexible mapreduced workflows

Data analytic applications are characterized by large data sets that are subject to a series of processing phases. Some of these phases are executed sequentially but others can be executed concurrently or in parallel on clusters, grids or clouds. The MapReduce programming model has been applied to process large data sets in cluster and cloud environments. For developing an application using MapReduce there is a need to install/configure/access specific frameworks such as Apache Hadoop or Elastic MapReduce in Amazon Cloud. It would be desirable to provide more flexibility in adjusting such configurations according to the application characteristics. Furthermore the composition of the multiple phases of a data analytic application requires the specification of all the phases and their orchestration. The original MapReduce model and environment lacks flexible support for such configuration and composition. Recognizing that scientific workflows have been successfully applied to modeling complex applications, this paper describes our experiments on implementing MapReduce as subworkflows in the AWARD framework (Autonomic Workflow Activities Reconfigurable and Dynamic). A text mining data analytic application is modeled as a complex workflow with multiple phases, where individual workflow nodes support MapReduce computations. As in typical MapReduce environments, the end user only needs to define the application algorithms for input data processing and for the map and reduce functions. In the paper we present experimental results when using the AWARD framework to execute MapReduce workflows deployed over multiple Amazon EC2 (Elastic Compute Cloud) instances.

El impacto de las aplicaciones intensivas de E/S en la planificación de trabajos en clusters no-dedicados

Con la mayor capacidad de los nodos de procesamiento en relación a la potencia de cómputo, cada vez más aplicaciones intensivas de datos como las aplicaciones de la bioinformática, se llevarán a ejecutar en clusters no dedicados. Los clusters no dedicados se caracterizan por su capacidad de combinar la ejecución de aplicaciones de usuarios locales con aplicaciones, científicas o comerciales, ejecutadas en paralelo. Saber qué efecto las aplicaciones con acceso intensivo a dados producen respecto a la mezcla de otro tipo (batch, interativa, SRT, etc) en los entornos no-dedicados permite el desarrollo de políticas de planificación más eficientes. Algunas de las aplicaciones intensivas de E/S se basan en el paradigma MapReduce donde los entornos que las utilizan, como Hadoop, se ocupan de la localidad de los datos, balanceo de carga de forma automática y trabajan con sistemas de archivos distribuidos. El rendimiento de Hadoop se puede mejorar sin aumentar los costos de hardware, al sintonizar varios parámetros de configuración claves para las especificaciones del cluster, para el tamaño de los datos de entrada y para el procesamiento complejo. La sincronización de estos parámetros de sincronización puede ser demasiado compleja para el usuario y/o administrador pero procura garantizar prestaciones más adecuadas. Este trabajo propone la evaluación del impacto de las aplicaciones intensivas de E/S en la planificación de trabajos en clusters no-dedicados bajo los paradigmas MPI y Mapreduce.

Análisis bioinformáticos sobre la tecnología Hadoop

Desde el inicio del proyecto del genoma humano y su éxito en el año 2001 se han secuenciado genomas de multitud de especies. La mejora en las tecnologías de secuenciación ha generado volúmenes de datos con un crecimiento exponencial. El proyecto Análisis bioinformáticos sobre la tecnología Hadoop abarca la computación paralela de datos biológicos como son las secuencias de ADN. El estudio ha sido encauzado por la naturaleza del problema a resolver. El alineamiento de secuencias genéticas con el paradigma MapReduce.

Integración de Hadoop con planificadores batch

Cada vez es mayor el número de aplicaciones desarrolladas en el ámbito científico, como en la Bioinformática o en las Geociencias, escritas bajo el modelo MapReduce, empleando herramientas de código abierto como Apache Hadoop. De la necesidad de integrar Hadoop en entornos HPC, para posibilitar la ejecutar aplicaciones desarrolladas bajo el paradigma MapReduce, nace el presente proyecto. Se analizan dos frameworks diseñados para facilitar dicha integración a los desarrolladores: HoD y myHadoop. En este proyecto se analiza, tanto las posibilidades en cuanto a entornos que ofrecen dichos frameworks para la ejecución de aplicaciones MapReduce, como el rendimiento de los clúster Hadoop generados con HoD o myHadoop respecto a un clúster Hadoop físico.

Otimização do processo de aprendizagem da estrutura gráfica de Redes Bayesianas em BigData

A automação na gestão e análise de dados tem sido um fator crucial para as empresas que necessitam de soluções eficientes em um mundo corporativo cada vez mais competitivo. A explosão do volume de informações, que vem se mantendo crescente nos últimos anos, tem exigido cada vez mais empenho em buscar estratégias para gerenciar e, principalmente, extrair informações estratégicas valiosas a partir do uso de algoritmos de Mineração de Dados, que comumente necessitam realizar buscas exaustivas na base de dados a fim de obter estatísticas que solucionem ou otimizem os parâmetros do modelo de extração do conhecimento utilizado; processo que requer computação intensiva para a execução de cálculos e acesso frequente à base de dados. Dada a eficiência no tratamento de incerteza, Redes Bayesianas têm sido amplamente utilizadas neste processo, entretanto, à medida que o volume de dados (registros e/ou atributos) aumenta, torna-se ainda mais custoso e demorado extrair informações relevantes em uma base de conhecimento. O foco deste trabalho é propor uma nova abordagem para otimização do aprendizado da estrutura da Rede Bayesiana no contexto de BigData, por meio do uso do processo de MapReduce, com vista na melhora do tempo de processamento. Para tanto, foi gerada uma nova metodologia que inclui a criação de uma Base de Dados Intermediária contendo todas as probabilidades necessárias para a realização dos cálculos da estrutura da rede. Por meio das análises apresentadas neste estudo, mostra-se que a combinação da metodologia proposta com o processo de MapReduce é uma boa alternativa para resolver o problema de escalabilidade nas etapas de busca em frequência do algoritmo K2 e, consequentemente, reduzir o tempo de resposta na geração da rede.

Managing Large Data Sets Using Support Vector Machines

Hundreds of Terabytes of CMS (Compact Muon Solenoid) data are being accumulated for storage day by day at the University of Nebraska-Lincoln, which is one of the eight US CMS Tier-2 sites. Managing this data includes retaining useful CMS data sets and clearing storage space for newly arriving data by deleting less useful data sets. This is an important task that is currently being done manually and it requires a large amount of time. The overall objective of this study was to develop a methodology to help identify the data sets to be deleted when there is a requirement for storage space. CMS data is stored using HDFS (Hadoop Distributed File System). HDFS logs give information regarding file access operations. Hadoop MapReduce was used to feed information in these logs to Support Vector Machines (SVMs), a machine learning algorithm applicable to classification and regression which is used in this Thesis to develop a classifier. Time elapsed in data set classification by this method is dependent on the size of the input HDFS log file since the algorithmic complexities of Hadoop MapReduce algorithms here are O(n). The SVM methodology produces a list of data sets for deletion along with their respective sizes. This methodology was also compared with a heuristic called Retention Cost which was calculated using size of the data set and the time since its last access to help decide how useful a data set is. Accuracies of both were compared by calculating the percentage of data sets predicted for deletion which were accessed at a later instance of time. Our methodology using SVMs proved to be more accurate than using the Retention Cost heuristic. This methodology could be used to solve similar problems involving other large data sets.

Una base dati per il knowledge discovery in genetica medica

L'innovazione delle tecnologie di sequenziamento negli ultimi anni ha reso possibile la catalogazione delle varianti genetiche nei campioni umani, portando nuove scoperte e comprensioni nella ricerca medica, farmaceutica, dell'evoluzione e negli studi sulla popolazione. La quantità di sequenze prodotta è molto cospicua, e per giungere all'identificazione delle varianti sono necessari diversi stadi di elaborazione delle informazioni genetiche in cui, ad ogni passo, vengono generate ulteriori informazioni. Insieme a questa immensa accumulazione di dati, è nata la necessità da parte della comunità scientifica di organizzare i dati in repository, dapprima solo per condividere i risultati delle ricerche, poi per permettere studi statistici direttamente sui dati genetici. Gli studi su larga scala coinvolgono quantità di dati nell'ordine dei petabyte, il cui mantenimento continua a rappresentare una sfida per le infrastrutture. Per la varietà e la quantità di dati prodotti, i database giocano un ruolo di primaria importanza in questa sfida. Modelli e organizzazione dei dati in questo campo possono fare la differenza non soltanto per la scalabilità, ma anche e soprattutto per la predisposizione al data mining. Infatti, la memorizzazione di questi dati in file con formati quasi-standard, la dimensione di questi file, e i requisiti computazionali richiesti, rendono difficile la scrittura di software di analisi efficienti e scoraggiano studi su larga scala e su dati eterogenei. Prima di progettare il database si è perciò studiata l’evoluzione, negli ultimi vent’anni, dei formati quasi-standard per i flat file biologici, contenenti metadati eterogenei e sequenze nucleotidiche vere e proprie, con record privi di relazioni strutturali. Recentemente questa evoluzione è culminata nell’utilizzo dello standard XML, ma i flat file delimitati continuano a essere gli standard più supportati da tools e piattaforme online. È seguita poi un’analisi dell’organizzazione interna dei dati per i database biologici pubblici. Queste basi di dati contengono geni, varianti genetiche, strutture proteiche, ontologie fenotipiche, relazioni tra malattie e geni, relazioni tra farmaci e geni. Tra i database pubblici studiati rientrano OMIM, Entrez, KEGG, UniProt, GO. L'obiettivo principale nello studio e nella modellazione del database genetico è stato quello di strutturare i dati in modo da integrare insieme i dati eterogenei prodotti e rendere computazionalmente possibili i processi di data mining. La scelta di tecnologia Hadoop/MapReduce risulta in questo caso particolarmente incisiva, per la scalabilità garantita e per l’efficienza nelle analisi statistiche più complesse e parallele, come quelle riguardanti le varianti alleliche multi-locus.