Systems operating in a medical environment need to maintain high standards regarding availability and performance. Large amounts of images are stored and studied to determine what is wrong with a patient. This puts hard requirements on the storage of the images. In this thesis, ways of incorporating distributed storage into a medical system are explored. Products, inspired by the success of Google, Amazon and others, are experimented with and compared to the current storage solutions. Several “non-relational databases” (NoSQL) are investigated for storing medically relevant metadata of images, while a set of distributed file systems are considered for storing the actual images. Distributed processing of the stored data is investigated by using Hadoop MapReduce to generate a useful model of the images' metadata.

The rapidly growing volume of electrophysiological signals has been generated for clinical research in neurological disorders. European Data Format (EDF) is a standard format for storing electrophysiological signals. However, the bottleneck of existing signal analysis tools for handling large-scale datasets is the sequential way of loading large EDF files before performing an analysis. To overcome this, we develop Hadoop-EDF, a distributed signal processing tool to load EDF data in a parallel manner using Hadoop MapReduce. Hadoop-EDF uses a robust data partition algorithm making EDF data parallel processable. We evaluate Hadoop-EDF’s scalability and performance by leveraging two datasets from the National Sleep Research Resource and running experiments on Amazon Web Service clusters. The performance of Hadoop-EDF on a 20-node cluster improves 27 times and 47 times than sequential processing of 200 small-size files and 200 large-size files, respectively. The results demonstrate that Hadoop-EDF is more suitable and effective in processing large EDF files.

Com o crescimento do volume de dados manipulado por aplicações de data warehousing, soluções centralizadas tornam-se muito custosas e enfrentam dificuldades para tratar a escalabilidade do volume de dados. Nesse sentido, existe a necessidade tanto de se armazenar grandes volumes de dados quanto de se realizar consultas analíticas (ou seja, consultas OLAP) sobre esses dados volumosos de forma eficiente. Isso pode ser facilitado por cenários caracterizados pelo uso de bancos de dados NoSQL gerenciados em ambientes paralelos e distribuídos. Dentre os desafios relacionados a esses cenários, destaca-se a necessidade de se promover uma análise de desempenho de aplicações de data warehousing que armazenam os dados do data warehouse (DW) em bancos de dados NoSQL orientados a colunas. A análise experimental e padronizada de diferentes sistemas é realizada por meio de ferramentas denominadas benchmarks. Entretanto, benchmarks para DW foram desenvolvidos majoritariamente para bancos de dados relacionais e ambientes centralizados. Nesta pesquisa de mestrado são investigadas formas de se estender o Star Schema Benchmark (SSB), um benchmark de DW centralizado, para o banco de dados NoSQL distribuído e orientado a colunas HBase. São realizadas propostas e análises principalmente baseadas em testes de desempenho experimentais considerando cada uma das quatro etapas de um benchmark, ou seja, esquema e carga de trabalho, geração de dados, parâmetros e métricas, e validação. Os principais resultados obtidos pelo desenvolvimento do trabalho são: (i) proposta do esquema FactDate, o qual otimiza consultas que acessam poucas dimensões do DW; (ii) investigação da aplicabilidade de diferentes esquemas a cenários empresariais distintos; (iii) proposta de duas consultas adicionais à carga de trabalho do SSB; (iv) análise da distribuição dos dados gerados pelo SSB, verificando se os dados agregados pelas consultas OLAP estão balanceados entre os nós de um cluster; (v) investigação da influência de três importantes parâmetros do framework Hadoop MapReduce no processamento de consultas OLAP; (vi) avaliação da relação entre o desempenho de consultas OLAP e a quantidade de nós que compõem um cluster; e (vii) proposta do uso de visões materializadas hierárquicas, por meio do framework Spark, para otimizar o desempenho no processamento de consultas OLAP consecutivas que requerem a análise de dados em níveis progressivamente mais ou menos detalhados. Os resultados obtidos representam descobertas importantes que visam possibilitar a proposta futura de um benchmark para DWs armazenados em bancos de dados NoSQL dentro de ambientes paralelos e distribuídos. / Due to the explosive increase in data volume, centralized data warehousing applications become very costly and are facing several problems to deal with data scalability. This is related to the fact that these applications need to store huge volumes of data and to perform analytical queries (i.e., OLAP queries) against these voluminous data efficiently. One solution is to employ scenarios characterized by the use of NoSQL databases managed in parallel and distributed environments. Among the challenges related to these scenarios, there is a need to investigate the performance of data warehousing applications that store the data warehouse (DW) in column-oriented NoSQL databases. In this context, benchmarks are widely used to perform standard and experimental analysis of distinct systems. However, most of the benchmarks for DW focus on relational database systems and centralized environments. In this masters research, we investigate how to extend the Star Schema Benchmark (SSB), which was proposed for centralized DWs, to the distributed and column-oriented NoSQL database HBase. We introduce proposals and analysis mainly based on experimental performance tests considering each one of the four steps of a benchmark, i.e. schema and workload, data generation, parameters and metrics, and validation. The main results described in this masters research are described as follows: (i) proposal of the FactDate schema, which optimizes queries that access few dimensions of the DW; (ii) investigation of the applicability of different schemas for different business scenarios; (iii) proposal of two additional queries to the SSB workload; (iv) analysis of the data distribution generated by the SSB, verifying if the data aggregated by OLAP queries are balanced between the nodes of a cluster; (v) investigation of the influence caused by three important parameters of the Hadoop MapReduce framework in the OLAP query processing; (vi) evaluation of the relationship between the OLAP query performance and the number of nodes of a cluster; and (vii) employment of hierarchical materialized views using the Spark framework to optimize the processing performance of consecutive OLAP queries that require progressively more or less aggregated data. These results represent important findings that enable the future proposal of a benchmark for DWs stored in NoSQL databases and managed in parallel and distributed environments.

Com o crescimento do volume de dados manipulado por aplicações de data warehousing, soluções centralizadas tornam-se muito custosas e enfrentam dificuldades para tratar a escalabilidade do volume de dados. Nesse sentido, existe a necessidade tanto de se armazenar grandes volumes de dados quanto de se realizar consultas analíticas (ou seja, consultas OLAP) sobre esses dados volumosos de forma eficiente. Isso pode ser facilitado por cenários caracterizados pelo uso de bancos de dados NoSQL gerenciados em ambientes paralelos e distribuídos. Dentre os desafios relacionados a esses cenários, destaca-se a necessidade de se promover uma análise de desempenho de aplicações de data warehousing que armazenam os dados do data warehouse (DW) em bancos de dados NoSQL orientados a colunas. A análise experimental e padronizada de diferentes sistemas é realizada por meio de ferramentas denominadas benchmarks. Entretanto, benchmarks para DW foram desenvolvidos majoritariamente para bancos de dados relacionais e ambientes centralizados. Nesta pesquisa de mestrado são investigadas formas de se estender o Star Schema Benchmark (SSB), um benchmark de DW centralizado, para o banco de dados NoSQL distribuído e orientado a colunas HBase. São realizadas propostas e análises principalmente baseadas em testes de desempenho experimentais considerando cada uma das quatro etapas de um benchmark, ou seja, esquema e carga de trabalho, geração de dados, parâmetros e métricas, e validação. Os principais resultados obtidos pelo desenvolvimento do trabalho são: (i) proposta do esquema FactDate, o qual otimiza consultas que acessam poucas dimensões do DW; (ii) investigação da aplicabilidade de diferentes esquemas a cenários empresariais distintos; (iii) proposta de duas consultas adicionais à carga de trabalho do SSB; (iv) análise da distribuição dos dados gerados pelo SSB, verificando se os dados agregados pelas consultas OLAP estão balanceados entre os nós de um cluster; (v) investigação da influência de três importantes parâmetros do framework Hadoop MapReduce no processamento de consultas OLAP; (vi) avaliação da relação entre o desempenho de consultas OLAP e a quantidade de nós que compõem um cluster; e (vii) proposta do uso de visões materializadas hierárquicas, por meio do framework Spark, para otimizar o desempenho no processamento de consultas OLAP consecutivas que requerem a análise de dados em níveis progressivamente mais ou menos detalhados. Os resultados obtidos representam descobertas importantes que visam possibilitar a proposta futura de um benchmark para DWs armazenados em bancos de dados NoSQL dentro de ambientes paralelos e distribuídos. / Due to the explosive increase in data volume, centralized data warehousing applications become very costly and are facing several problems to deal with data scalability. This is related to the fact that these applications need to store huge volumes of data and to perform analytical queries (i.e., OLAP queries) against these voluminous data efficiently. One solution is to employ scenarios characterized by the use of NoSQL databases managed in parallel and distributed environments. Among the challenges related to these scenarios, there is a need to investigate the performance of data warehousing applications that store the data warehouse (DW) in column-oriented NoSQL databases. In this context, benchmarks are widely used to perform standard and experimental analysis of distinct systems. However, most of the benchmarks for DW focus on relational database systems and centralized environments. In this masters research, we investigate how to extend the Star Schema Benchmark (SSB), which was proposed for centralized DWs, to the distributed and column-oriented NoSQL database HBase. We introduce proposals and analysis mainly based on experimental performance tests considering each one of the four steps of a benchmark, i.e. schema and workload, data generation, parameters and metrics, and validation. The main results described in this masters research are described as follows: (i) proposal of the FactDate schema, which optimizes queries that access few dimensions of the DW; (ii) investigation of the applicability of different schemas for different business scenarios; (iii) proposal of two additional queries to the SSB workload; (iv) analysis of the data distribution generated by the SSB, verifying if the data aggregated by OLAP queries are balanced between the nodes of a cluster; (v) investigation of the influence caused by three important parameters of the Hadoop MapReduce framework in the OLAP query processing; (vi) evaluation of the relationship between the OLAP query performance and the number of nodes of a cluster; and (vii) employment of hierarchical materialized views using the Spark framework to optimize the processing performance of consecutive OLAP queries that require progressively more or less aggregated data. These results represent important findings that enable the future proposal of a benchmark for DWs stored in NoSQL databases and managed in parallel and distributed environments.

Master of Science / Department of Computing and Information Sciences / Mitchell L. Neilsen / Recently, cost-effective and timely processing of large datasets has been playing an important role in the success of many enterprises and the scientific computing community. Two promising trends ensure that applications will be able to deal with ever increasing data volumes: first, the emergence of cloud computing, which provides transparent access to a large number of processing, storage and networking resources; and second, the development of the MapReduce programming model, which provides a high-level abstraction for data-intensive computing. MapReduce has been widely used for large-scale data analysis in the Cloud [5]. The system is well recognized for its elastic scalability and fine-grained fault tolerance. However, even to run a single program in a MapReduce framework, a number of tuning parameters have to be set by users or system administrators to increase the efficiency of the program. Users often run into performance problems because they are unaware of how to set these parameters, or because they don't even know that these parameters exist. With MapReduce being a relatively new technology, it is not easy to find qualified administrators [4]. The major objective of this project is to provide a framework that optimizes MapReduce programs that run on large datasets. This is done by executing the MapReduce program on a part of the dataset using stored parameter combinations and setting the program with the most efficient combination and this modified program can be executed over the different datasets. We know that many MapReduce programs are used over and over again in applications like daily weather analysis, log analysis, daily report generation etc. So, once the parameter combination is set, it can be used on a number of data sets efficiently. This feature can go a long way towards improving the productivity of users who lack the skills to optimize programs themselves due to lack of familiarity with MapReduce or with the data being processed.

In the last decade, advances in sensor and computing technology are revolutionary. The latest-generation of hyperspectral and synthetic aperture radar ((SAR) instruments have increased their spectral, spatial, and temporal resolution. Consequently, the data sets collected are increasing rapidly in size and frequency of acquisition. Remote sensing applications are requiring more computing resources for data analysis. High performance computing (HPC) infrastructure such as clusters, distributed networks, grids, clouds and specialized hardware components, have been used to disseminate large volumes of remote sensing data and to accelerate the computational speed in processing raw images and extracting information from remote sensing data. In previous research we have shown that we can improve computational efficiency of a hyperspectral image denoising algorithm by parallelizing the algorithm utilizing a distributed computing grid. In recent years, computational cloud technology is emerging, bringing more flexibility and simplicity for data processing. Hadoop MapReduce is a software framework for distributed commodity computing clusters, allowing parallel processing of massive datasets. In this project, we implement a software application to map forest aboveground biomass (AGB) with normalized difference vegetation indices (NDVI) using Landsat Thematic Mapper’s bands 4 and 5 (ND45). We present observations and experimental results on the performance and the algorithmic complexity of the implementation. There are three research questions answered in this thesis, as follows. 1) How do we implement remote sensing algorithms, such as forest AGB mapping, in a computer cloud environment? 2) What are the requirements to implement distributed processing of remote sensing images using the cloud programming model? 3) What is the performance increase for large area remote sensing image processing in a cloud environment? / Graduate / 0799 / 0984

This thesis deals with the subject of distributed processing of IP flow. Main goal is to provide an implementation of a software collector which allows storing and processing huge amount of a network data in particular. There was studied an open-source implementation of a framework for the distributed processing of large data sets called Hadoop, which is based on MapReduce paradigm. There were made some experiments with this system which provided the comparison with the current systems and shown weaknesses of this framework. Based on this knowledge there was created a specification and scheme for an extension of current software collector within this work. In terms of the created scheme there was created an implementation of query framework for formed collector, which is considered as most critical in the field of distributed processing of IP flow data. Results of experiments with created implementation show significant performance growth and ability of linear scalability with some types of queries.

Im Bereich von Entity-Resolution oder deduplication werden aufgrund fehlender global eindeutiger Identifikatoren Match-Techniken verwendet, um zu bestimmen, ob verschiedene Datensätze dasselbe Realweltobjekt darstellen. Die inhärente quadratische Komplexität führt zu sehr langen Laufzeiten für große Datenmengen, was eine Parallelisierung dieses Prozesses erfordert. MapReduce ist wegen seiner Skalierbarkeit und Einsetzbarkeit in Cloud- Infrastrukturen eine gute Lösung zur Verbesserung der Laufzeit. Außerdem kann unter bestimmten Voraussetzungen die Qualität des Match-Ergebnisses durch die Berechnung der transitiven Hülle verbessert werden.

Les workflows sont des systèmes typiques traitant le big data. Ces systèmes sont déployés sur des sites géo-distribués pour exploiter des infrastructures cloud existantes et réaliser des expériences à grande échelle. Les données générées par de telles expériences sont considérables et stockées à plusieurs endroits pour être réutilisées. En effet, les systèmes workflow sont composés de tâches collaboratives, présentant de nouveaux besoins en terme de dépendance et d'échange de données intermédiaires pour leur traitement. Cela entraîne de nouveaux problèmes lors de la sélection de données distribuées et de ressources de stockage, de sorte que l'exécution des tâches ou du job s'effectue à temps et que l'utilisation des ressources soit rentable. Par conséquent, cette thèse aborde le problème de gestion des données hébergées dans des centres de données cloud en considérant les exigences des systèmes workflow qui les génèrent. Pour ce faire, le premier problème abordé dans cette thèse traite le comportement d'accès aux données intermédiaires des tâches qui sont exécutées dans un cluster MapReduce-Hadoop. Cette approche développe et explore le modèle de Markov qui utilise la localisation spatiale des blocs et analyse la séquentialité des fichiers spill à travers un modèle de prédiction. Deuxièmement, cette thèse traite le problème de placement de données intermédiaire dans un stockage cloud fédéré en minimisant le coût de stockage. A travers les mécanismes de fédération, nous proposons un algorithme exacte ILP afin d’assister plusieurs centres de données cloud hébergeant les données de dépendances en considérant chaque paire de fichiers. Enfin, un problème plus générique est abordé impliquant deux variantes du problème de placement lié aux dépendances divisibles et entières. L'objectif principal est de minimiser le coût opérationnel en fonction des besoins de dépendances inter et intra-job / The typical cloud big data systems are the workflow-based including MapReduce which has emerged as the paradigm of choice for developing large scale data intensive applications. Data generated by such systems are huge, valuable and stored at multiple geographical locations for reuse. Indeed, workflow systems, composed of jobs using collaborative task-based models, present new dependency and intermediate data exchange needs. This gives rise to new issues when selecting distributed data and storage resources so that the execution of tasks or job is on time, and resource usage-cost-efficient. Furthermore, the performance of the tasks processing is governed by the efficiency of the intermediate data management. In this thesis we tackle the problem of intermediate data management in cloud multi-datacenters by considering the requirements of the workflow applications generating them. For this aim, we design and develop models and algorithms for big data placement problem in the underlying geo-distributed cloud infrastructure so that the data management cost of these applications is minimized. The first addressed problem is the study of the intermediate data access behavior of tasks running in MapReduce-Hadoop cluster. Our approach develops and explores Markov model that uses spatial locality of intermediate data blocks and analyzes spill file sequentiality through a prediction algorithm. Secondly, this thesis deals with storage cost minimization of intermediate data placement in federated cloud storage. Through a federation mechanism, we propose an exact ILP algorithm to assist multiple cloud datacenters hosting the generated intermediate data dependencies of pair of files. The proposed algorithm takes into account scientific user requirements, data dependency and data size. Finally, a more generic problem is addressed in this thesis that involve two variants of the placement problem: splittable and unsplittable intermediate data dependencies. The main goal is to minimize the operational data cost according to inter and intra-job dependencies

With the growing business impact of distributed big data analytics jobs, it has become crucial to optimize their execution and resource consumption. In most cases, such jobs consist of multiple sub-entities called tasks and are executed online in a large shared distributed computing system. The ability to accurately estimate runtime properties and coordinate execution of sub-entities of a job allows a scheduler to efficiently schedule jobs for optimal scheduling. This thesis presents the first study that highlights spatial dimension, an inherent property of distributed jobs, and underscores its importance in efficient cluster job scheduling. We develop two new classes of spatial dimension based algorithms to<br>address the two primary challenges of cluster scheduling. First, we propose, validate, and design two complete systems that employ learning algorithms exploiting spatial dimension. We demonstrate high similarity in runtime properties between sub-entities of the same job by detailed trace analysis on four different industrial cluster traces. We identify design challenges and propose principles for a sampling based learning system for two examples, first for a coflow scheduler, and second for a cluster job scheduler.<br>We also propose, design, and demonstrate the effectiveness of new multi-task scheduling algorithms based on effective synchronization across the spatial dimension. We underline and validate by experimental analysis the importance of synchronization between sub-entities (flows, tasks) of a distributed entity (coflow, data analytics jobs) for its efficient execution. We also highlight that by not considering sibling sub-entities when scheduling something it may also lead to sub-optimal overall cluster performance. We propose, design, and implement a full coflow scheduler based on these assertions.