Harmony: Towards automated self-adaptive consistency in cloud storage

In just a few years cloud computing has become a very popular paradigm and a business success story, with storage being one of the key features. To achieve high data availability, cloud storage services rely on replication. In this context, one major challenge is data consistency. In contrast to traditional approaches that are mostly based on strong consistency, many cloud storage services opt for weaker consistency models in order to achieve better availability and performance. This comes at the cost of a high probability of stale data being read, as the replicas involved in the reads may not always have the most recent write. In this paper, we propose a novel approach, named Harmony, which adaptively tunes the consistency level at run-time according to the application requirements. The key idea behind Harmony is an intelligent estimation model of stale reads, allowing to elastically scale up or down the number of replicas involved in read operations to maintain a low (possibly zero) tolerable fraction of stale reads. As a result, Harmony can meet the desired consistency of the applications while achieving good performance. We have implemented Harmony and performed extensive evaluations with the Cassandra cloud storage on Grid?5000 testbed and on Amazon EC2. The results show that Harmony can achieve good performance without exceeding the tolerated number of stale reads. For instance, in contrast to the static eventual consistency used in Cassandra, Harmony reduces the stale data being read by almost 80% while adding only minimal latency. Meanwhile, it improves the throughput of the system by 45% while maintaining the desired consistency requirements of the applications when compared to the strong consistency model in Cassandra.

Sistema de monitorización de sistemas distribuidos

La computación distribuida ha estado presente desde hace unos cuantos años, pero es quizás en la actualidad cuando está contando con una mayor repercusión. En los últimos años el modelo de computación en la nube (Cloud computing) ha ganado mucha popularidad, prueba de ello es la cantidad de productos existentes. Todo sistema informático requiere ser controlado a través de sistemas de monitorización que permiten conocer el estado del mismo, de tal manera que pueda ser gestionado fácilmente. Hoy en día la mayoría de los productos de monitorización existentes limitan a la hora de visualizar una representación real de la arquitectura de los sistemas a monitorizar, lo que puede dificultar la tarea de los administradores. Es decir, la visualización que proporcionan de la arquitectura del sistema, en muchos casos se ve influenciada por el diseño del sistema de visualización, lo que impide ver los niveles de la arquitectura y las relaciones entre estos. En este trabajo se presenta un sistema de monitorización para sistemas distribuidos o Cloud, que pretende dar solución a esta problemática, no limitando la representación de la arquitectura del sistema a monitorizar. El sistema está formado por: agentes, que se encargan de la tarea de recolección de las métricas del sistema monitorizado; un servidor, al que los agentes le envían las métricas para que las almacenen en una base de datos; y una aplicación web, a través de la que se visualiza toda la información. El sistema ha sido probado satisfactoriamente con la monitorización de CumuloNimbo, una plataforma como servicio (PaaS), que ofrece interfaz SQL y procesamiento transaccional altamente escalable sobre almacenes clave valor. Este trabajo describe la arquitectura del sistema de monitorización, y en concreto, el desarrollo de la principal contribución al sistema, la aplicación web. ---ABSTRACT---Distributed computing has been around for quite a long time, but now it is becoming more and more important. In the last few years, cloud computing, a branch of distributed computing has become very popular, as its different products in the market can prove. Every computing system requires to be controlled through monitoring systems to keep them functioning correctly. Currently, most of the monitoring systems in the market only provide a view of the architectures of the systems monitored, which in most cases do not permit having a real view of the system. This lack of vision can make administrators’ tasks really difficult. If they do not know the architecture perfectly, controlling the system based on the view that the monitoring system provides is extremely complicated. The project introduces a new monitoring system for distributed or Cloud systems, which shows the real architecture of the system. This new system is composed of several elements: agents, which collect the metrics of the monitored system; a server, which receives the metrics from the agents and saves them in a database; and a web application, which shows all the data collected in an easy way. The monitoring system has been tested successfully with Cumulonimbo. CumuloNimbo is a platform as a service (PaaS) which offers an SQL interface and a high-scalable transactional process. This platform works over key-value storage. This project describes the architecture of the monitoring system, especially, the development of the web application, which is the main contribution to the system.

A proposal for a modular and application-aware autonomic manager of private cloud infrastructures

Recientemente, el paradigma de la computación en la nube ha recibido mucho interés por parte tanto de la industria como del mundo académico. Las infraestructuras cloud públicas están posibilitando nuevos modelos de negocio y ayudando a reducir costes. Sin embargo, una compañía podría desear ubicar sus datos y servicios en sus propias instalaciones, o tener que atenerse a leyes de protección de datos. Estas circunstancias hacen a las infraestructuras cloud privadas ciertamente deseables, ya sea para complementar a las públicas o para sustituirlas por completo. Por desgracia, las carencias en materia de estándares han impedido que las soluciones para la gestión de infraestructuras privadas se hayan desarrollado adecuadamente. Además, la multitud de opciones disponibles ha creado en los clientes el miedo a depender de una tecnología concreta (technology lock-in). Una de las causas de este problema es la falta de alineación entre la investigación académica y los productos comerciales, ya que aquella está centrada en el estudio de escenarios idealizados sin correspondencia con el mundo real, mientras que éstos consisten en soluciones desarrolladas sin tener en cuenta cómo van a encajar con los estándares más comunes o sin preocuparse de hacer públicos sus resultados. Con objeto de resolver este problema, propongo un sistema de gestión modular para infraestructuras cloud privadas enfocado en tratar con las aplicaciones en lugar de centrarse únicamente en los recursos hardware. Este sistema de gestión sigue el paradigma de la computación autónoma y está diseñado en torno a un modelo de información sencillo, desarrollado para ser compatible con los estándares más comunes. Este modelo divide el entorno en dos vistas, que sirven para separar aquello que debe preocupar a cada actor involucrado del resto de información, pero al mismo tiempo permitiendo relacionar el entorno físico con las máquinas virtuales que se despliegan encima de él. En dicho modelo, las aplicaciones cloud están divididas en tres tipos genéricos (Servicios, Trabajos de Big Data y Reservas de Instancias), para que así el sistema de gestión pueda sacar partido de las características propias de cada tipo. El modelo de información está complementado por un conjunto de acciones de gestión atómicas, reversibles e independientes, que determinan las operaciones que se pueden llevar a cabo sobre el entorno y que es usado para hacer posible la escalabilidad en el entorno. También describo un motor de gestión encargado de, a partir del estado del entorno y usando el ya mencionado conjunto de acciones, la colocación de recursos. Está dividido en dos niveles: la capa de Gestores de Aplicación, encargada de tratar sólo con las aplicaciones; y la capa del Gestor de Infraestructura, responsable de los recursos físicos. Dicho motor de gestión obedece un ciclo de vida con dos fases, para así modelar mejor el comportamiento de una infraestructura real. El problema de la colocación de recursos es atacado durante una de las fases (la de consolidación) por un resolutor de programación entera, y durante la otra (la online) por un heurístico hecho ex-profeso. Varias pruebas han demostrado que este acercamiento combinado es superior a otras estrategias. Para terminar, el sistema de gestión está acoplado a arquitecturas de monitorización y de actuadores. Aquella estando encargada de recolectar información del entorno, y ésta siendo modular en su diseño y capaz de conectarse con varias tecnologías y ofrecer varios modos de acceso. ABSTRACT The cloud computing paradigm has raised in popularity within the industry and the academia. Public cloud infrastructures are enabling new business models and helping to reduce costs. However, the desire to host company’s data and services on premises, and the need to abide to data protection laws, make private cloud infrastructures desirable, either to complement or even fully substitute public oferings. Unfortunately, a lack of standardization has precluded private infrastructure management solutions to be developed to a certain level, and a myriad of diferent options have induced the fear of lock-in in customers. One of the causes of this problem is the misalignment between academic research and industry ofering, with the former focusing in studying idealized scenarios dissimilar from real-world situations, and the latter developing solutions without taking care about how they f t with common standards, or even not disseminating their results. With the aim to solve this problem I propose a modular management system for private cloud infrastructures that is focused on the applications instead of just the hardware resources. This management system follows the autonomic system paradigm, and is designed around a simple information model developed to be compatible with common standards. This model splits the environment in two views that serve to separate the concerns of the stakeholders while at the same time enabling the traceability between the physical environment and the virtual machines deployed onto it. In it, cloud applications are classifed in three broad types (Services, Big Data Jobs and Instance Reservations), in order for the management system to take advantage of each type’s features. The information model is paired with a set of atomic, reversible and independent management actions which determine the operations that can be performed over the environment and is used to realize the cloud environment’s scalability. From the environment’s state and using the aforementioned set of actions, I also describe a management engine tasked with the resource placement. It is divided in two tiers: the Application Managers layer, concerned just with applications; and the Infrastructure Manager layer, responsible of the actual physical resources. This management engine follows a lifecycle with two phases, to better model the behavior of a real infrastructure. The placement problem is tackled during one phase (consolidation) by using an integer programming solver, and during the other (online) with a custom heuristic. Tests have demonstrated that this combined approach is superior to other strategies. Finally, the management system is paired with monitoring and actuators architectures. The former able to collect the necessary information from the environment, and the later modular in design and capable of interfacing with several technologies and ofering several access interfaces.

Heterogeneous architecture for the optimization of large-scale graph processing data centres

Con el auge del Cloud Computing, las aplicaciones de proceso de datos han sufrido un incremento de demanda, y por ello ha cobrado importancia lograr m�ás eficiencia en los Centros de Proceso de datos. El objetivo de este trabajo es la obtenci�ón de herramientas que permitan analizar la viabilidad y rentabilidad de diseñar Centros de Datos especializados para procesamiento de datos, con una arquitectura, sistemas de refrigeraci�ón, etc. adaptados. Algunas aplicaciones de procesamiento de datos se benefician de las arquitecturas software, mientras que en otras puede ser m�ás eficiente un procesamiento con arquitectura hardware. Debido a que ya hay software con muy buenos resultados en el procesamiento de grafos, como el sistema XPregel, en este proyecto se realizará una arquitectura hardware en VHDL, implementando el algoritmo PageRank de Google de forma escalable. Se ha escogido este algoritmo ya que podr��á ser m�ás eficiente en arquitectura hardware, debido a sus características concretas que se indicaráan m�ás adelante. PageRank sirve para ordenar las p�áginas por su relevancia en la web, utilizando para ello la teorí��a de grafos, siendo cada página web un vértice de un grafo; y los enlaces entre páginas, las aristas del citado grafo. En este proyecto, primero se realizará un an�álisis del estado de la técnica. Se supone que la implementaci�ón en XPregel, un sistema de procesamiento de grafos, es una de las m�ás eficientes. Por ello se estudiará esta �ultima implementaci�ón. Sin embargo, debido a que Xpregel procesa, en general, algoritmos que trabajan con grafos; no tiene en cuenta ciertas caracterí��sticas del algoritmo PageRank, por lo que la implementaci�on no es �optima. Esto es debido a que en PageRank, almacenar todos los datos que manda un mismo v�értice es un gasto innecesario de memoria ya que todos los mensajes que manda un vértice son iguales entre sí e iguales a su PageRank. Se realizará el diseño en VHDL teniendo en cuenta esta caracter��ística del citado algoritmo,evitando almacenar varias veces los mensajes que son iguales. Se ha elegido implementar PageRank en VHDL porque actualmente las arquitecturas de los sistemas operativos no escalan adecuadamente. Se busca evaluar si con otra arquitectura se obtienen mejores resultados. Se realizará un diseño partiendo de cero, utilizando la memoria ROM de IPcore de Xillinx (Software de desarrollo en VHDL), generada autom�áticamente. Se considera hacer cuatro tipos de módulos para que as�� el procesamiento se pueda hacer en paralelo. Se simplificar�á la estructura de XPregel con el fin de intentar aprovechar la particularidad de PageRank mencionada, que hace que XPregel no le saque el m�aximo partido. Despu�és se escribirá el c�ódigo, realizando una estructura escalable, ya que en la computación intervienen millones de páginas web. A continuación, se sintetizar�á y se probará el código en una FPGA. El �ultimo paso será una evaluaci�ón de la implementaci�ón, y de posibles mejoras en cuanto al consumo.

Implementación de un sistema informático para el procesamiento de datos de tráfico vial

El presente Trabajo de Fin de Grado se enmarca dentro de un sistema de control y desarrollo de sistemas inteligentes de transporte (ITS). Este Trabajo consta de varias líneas de desarrollo, que se engloban dentro de dicho marco y surgen de la necesidad de aumentar la seguridad, flujo, estructura y mantenimiento de las carreteras incorporando las tecnologías más recientes. En primer lugar, el presente Trabajo se centra en el desarrollo de un nuevo sistema de procesamiento de datos de tráfico en tiempo real que aprovecha las tecnologías de Big Data, Cloud Computing y Map-Reduce que han surgido estos últimos años. Para ello se realiza un estudio previo de los datos de tráfico vial que originan los vehículos que viajan por carreteras. Centrándose en el sistema empleado por la Dirección General de Tráfico de España y comparándolos con el de las Empresas basadas en servicios de localización (LBS). Se expone el modelo Hadoop utilizado así como el proceso Map-Reduce implementado en este sistema analizador. Por último los datos de salida son preparados y enviados a un módulo web básico que actúa como Sistema de Información Geográfica (GIS).---ABSTRACT---This Final Degree Project is part of a control system and development of intelligent transport systems (ITS). This work is part of a several lines of development, which are included within this framework and arise from the need to increase security, flow, structure and maintenance of roads incorporating the latest technologies. First, this paper focuses on the development of a new data processing system of real-time traffic that takes advantage of Big Data, Cloud Computing and Map-Reduce technologies emerged in our recent years. It is made a preliminary study of road traffic data originated by vehicles traveling by road. Focusing on the system used by the Dirección General de Tráfico of Spain and compared with that of the companies offering location based services (LBS). It is exposed the used Hadoop model and the Map-Reduce process implemented on this analyzer system. Finally, the output data is prepared and sent to a basic web module that acts as Geographic Information System (GIS).

Tecnologie per l’analisi in tempo reale di Big Data: prestazioni a confronto

Lo scopo di questo l'elaborato è l'analisi,lo studio e il confronto delle tecnologie per l'analisi in tempo reale di Big Data: Apache Spark Streaming, Apache Storm e Apache Flink. Per eseguire un adeguato confronto si è deciso di realizzare un sistema di rilevamento e riconoscimento facciale all’interno di un video, in maniera da poter parallelizzare le elaborazioni necessarie sfruttando le potenzialità di ogni architettura. Dopo aver realizzato dei prototipi realistici, uno per ogni architettura, si è passati alla fase di testing per misurarne le prestazioni. Attraverso l’impiego di cluster appositamente realizzati in ambiente locale e cloud, sono state misurare le caratteristiche che rappresentavano, meglio di altre, le differenze tra le architetture, cercando di dimostrarne quantitativamente l’efficacia degli algoritmi utilizzati e l’efficienza delle stesse. Si è scelto quindi il massimo input rate sostenibile e la latenza misurate al variare del numero di nodi. In questo modo era possibile osservare la scalabilità di architettura, per analizzarne l’andamento e verificare fino a che limite si potesse giungere per mantenere un compromesso accettabile tra il numero di nodi e l’input rate sostenibile. Gli esperimenti effettuati hanno mostrato che, all’aumentare del numero di worker le prestazioni del sistema migliorano, rendendo i sistemi studiati adatti all’utilizzo su larga scala. Inoltre sono state rilevate sostanziali differenze tra i vari framework, riportando pro e contro di ognuno, cercando di evidenziarne i più idonei al caso di studio.

Cumulon: Simplified Matrix-Based Data Analytics in the Cloud

Cumulon is a system aimed at simplifying the development and deployment of statistical analysis of big data in public clouds. Cumulon allows users to program in their familiar language of matrices and linear algebra, without worrying about how to map data and computation to specific hardware and cloud software platforms. Given user-specified requirements in terms of time, monetary cost, and risk tolerance, Cumulon automatically makes intelligent decisions on implementation alternatives, execution parameters, as well as hardware provisioning and configuration settings -- such as what type of machines and how many of them to acquire. Cumulon also supports clouds with auction-based markets: it effectively utilizes computing resources whose availability varies according to market conditions, and suggests best bidding strategies for them. Cumulon explores two alternative approaches toward supporting such markets, with different trade-offs between system and optimization complexity. Experimental study is conducted to show the efficiency of Cumulon's execution engine, as well as the optimizer's effectiveness in finding the optimal plan in the vast plan space.

Studies of CMS data access patterns with machine learning techniques

This thesis presents a study of the Grid data access patterns in distributed analysis in the CMS experiment at the LHC accelerator. This study ranges from the deep analysis of the historical patterns of access to the most relevant data types in CMS, to the exploitation of a supervised Machine Learning classification system to set-up a machinery able to eventually predict future data access patterns - i.e. the so-called dataset “popularity” of the CMS datasets on the Grid - with focus on specific data types. All the CMS workflows run on the Worldwide LHC Computing Grid (WCG) computing centers (Tiers), and in particular the distributed analysis systems sustains hundreds of users and applications submitted every day. These applications (or “jobs”) access different data types hosted on disk storage systems at a large set of WLCG Tiers. The detailed study of how this data is accessed, in terms of data types, hosting Tiers, and different time periods, allows to gain precious insight on storage occupancy over time and different access patterns, and ultimately to extract suggested actions based on this information (e.g. targetted disk clean-up and/or data replication). In this sense, the application of Machine Learning techniques allows to learn from past data and to gain predictability potential for the future CMS data access patterns. Chapter 1 provides an introduction to High Energy Physics at the LHC. Chapter 2 describes the CMS Computing Model, with special focus on the data management sector, also discussing the concept of dataset popularity. Chapter 3 describes the study of CMS data access patterns with different depth levels. Chapter 4 offers a brief introduction to basic machine learning concepts and gives an introduction to its application in CMS and discuss the results obtained by using this approach in the context of this thesis.

Machine learning as a service for high energy physics (MLaaS4HEP): a service for ML-based data analyses

With the CERN LHC program underway, there has been an acceleration of data growth in the High Energy Physics (HEP) field and the usage of Machine Learning (ML) in HEP will be critical during the HL-LHC program when the data that will be produced will reach the exascale. ML techniques have been successfully used in many areas of HEP nevertheless, the development of a ML project and its implementation for production use is a highly time-consuming task and requires specific skills. Complicating this scenario is the fact that HEP data is stored in ROOT data format, which is mostly unknown outside of the HEP community. The work presented in this thesis is focused on the development of a ML as a Service (MLaaS) solution for HEP, aiming to provide a cloud service that allows HEP users to run ML pipelines via HTTP calls. These pipelines are executed by using the MLaaS4HEP framework, which allows reading data, processing data, and training ML models directly using ROOT files of arbitrary size from local or distributed data sources. Such a solution provides HEP users non-expert in ML with a tool that allows them to apply ML techniques in their analyses in a streamlined manner. Over the years the MLaaS4HEP framework has been developed, validated, and tested and new features have been added. A first MLaaS solution has been developed by automatizing the deployment of a platform equipped with the MLaaS4HEP framework. Then, a service with APIs has been developed, so that a user after being authenticated and authorized can submit MLaaS4HEP workflows producing trained ML models ready for the inference phase. A working prototype of this service is currently running on a virtual machine of INFN-Cloud and is compliant to be added to the INFN Cloud portfolio of services.

Android as a cloud ticket validator

Trabalho Final de Mestrado para a obtenção do grau de Mestre em Engenharia Informática e de Computadores

Cloud-assisted Wireless Body Area Networks

Wireless Body Area Networks (WBANs) have emerged as a promising technology for medical and non-medical applications. WBANs consist of a number of miniaturized, portable, and autonomous sensor nodes that are used for long-term health monitoring of patients. These sensor nodes continuously collect information of patients, which are used for ubiquitous health monitoring. In addition, WBANs may be used for managing catastrophic events and increasing the effectiveness and performance of rescue forces. The huge amount of data collected by WBAN nodes demands scalable, on-demand, powerful, and secure storage and processing infrastructure. Cloud computing is expected to play a significant role in achieving the aforementioned objectives. The cloud computing environment links different devices ranging from miniaturized sensor nodes to high-performance supercomputers for delivering people-centric and context-centric services to the individuals and industries. The possible integration of WBANs with cloud computing (WBAN-cloud) will introduce viable and hybrid platform that must be able to process the huge amount of data collected from multiple WBANs. This WBAN-cloud will enable users (including physicians and nurses) to globally access the processing and storage infrastructure at competitive costs. Because WBANs forward useful and life-critical information to the cloud – which may operate in distributed and hostile environments, novel security mechanisms are required to prevent malicious interactions to the storage infrastructure. Both the cloud providers and the users must take strong security measures to protect the storage infrastructure.

Middleware for Large-scale Distributed Systems

Nos últimos anos o aumento exponencial da utilização de dispositivos móveis e serviços disponibilizados na “Cloud” levou a que a forma como os sistemas são desenhados e implementados mudasse, numa perspectiva de tentar alcançar requisitos que até então não eram essenciais. Analisando esta evolução, com o enorme aumento dos dispositivos móveis, como os “smartphones” e “tablets” fez com que o desenho e implementação de sistemas distribuidos fossem ainda mais importantes nesta área, na tentativa de promover sistemas e aplicações que fossem mais flexíveis, robutos, escaláveis e acima de tudo interoperáveis. A menor capacidade de processamento ou armazenamento destes dispositivos tornou essencial o aparecimento e crescimento de tecnologias que prometem solucionar muitos dos problemas identificados. O aparecimento do conceito de Middleware visa solucionar estas lacunas nos sistemas distribuidos mais evoluídos, promovendo uma solução a nível de organização e desenho da arquitetura dos sistemas, ao memo tempo que fornece comunicações extremamente rápidas, seguras e de confiança. Uma arquitetura baseada em Middleware visa dotar os sistemas de um canal de comunicação que fornece uma forte interoperabilidade, escalabilidade, e segurança na troca de mensagens, entre outras vantagens. Nesta tese vários tipos e exemplos de sistemas distribuídos e são descritos e analisados, assim como uma descrição em detalhe de três protocolos (XMPP, AMQP e DDS) de comunicação, sendo dois deles (XMPP e AMQP) utilzados em projecto reais que serão descritos ao longo desta tese. O principal objetivo da escrita desta tese é demonstrar o estudo e o levantamento do estado da arte relativamente ao conceito de Middleware aplicado a sistemas distribuídos de larga escala, provando que a utilização de um Middleware pode facilitar e agilizar o desenho e desenvolvimento de um sistema distribuído e traz enormes vantagens num futuro próximo.

TSKY: a dependable middleware solution for data privacy using public storage clouds

Dissertação para obtenção do Grau de Mestre em Engenharia Informática

Pervasive data mining engine

Dissertação de mestrado integrado em Engenharia e Gestão de Sistemas de Informação

Communications with QoS in IoT&Cloud environment

Dissertação de mestrado integrado em Engenharia de Telecomunicações e Informática