• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 2
  • Tagged with
  • 3
  • 3
  • 2
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • 1
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
1

A Performance Comparison of VMware GPU Virtualization Techniques in Cloud Gaming

2016 March 1900 (has links)
Cloud gaming is an application deployment scenario which runs an interactive gaming application remotely in a cloud according to the commands received from a thin client and streams the scenes as a video sequence back to the client over the Internet, and it is of interest to both research community and industry. The academic community has developed some open-source cloud gaming systems such as GamingAnywhere for research study, while some industrial pioneers such as Onlive and Gaikai have succeeded in gaining a large user base in the cloud gaming market. Graphical Processing Unit (GPU) virtualization plays an important role in such an environment as it is a critical component that allows virtual machines to run 3D applications with performance guarantees. Currently, GPU pass-through and GPU sharing are the two main techniques of GPU virtualization. The former enables a single virtual machine to access a physical GPU directly and exclusively, while the latter makes a physical GPU shareable by multiple virtual machines. VMware Inc., one of the most popular virtualization solution vendors, has provided concrete implementations of GPU pass-through and GPU sharing. In particular, it provides a GPU pass-through solution called Virtual Dedicated Graphics Acceleration (vDGA) and a GPU-sharing solution called Virtual Shared Graphics Acceleration (vSGA). Moreover, VMware Inc. recently claimed it realized another GPU sharing solution called vGPU. Nevertheless, the feasibility and performance of these solutions in cloud gaming has not been studied yet. In this work, an experimental study is conducted to evaluate the feasibility and performance of GPU pass-through and GPU sharing solutions offered by VMware in cloud gaming scenarios. The primary results confirm that vDGA and vGPU techniques can fit the demands of cloud gaming. In particular, these two solutions achieved good performance in the tested graphics card benchmarks, and gained acceptable image quality and response delay for the tested games.
2

Generalizing the Utility of Graphics Processing Units in Large-Scale Heterogeneous Computing Systems

Xiao, Shucai 03 July 2013 (has links)
Today, heterogeneous computing systems are widely used to meet the increasing demand for high-performance computing. These systems commonly use powerful and energy-efficient accelerators to augment general-purpose processors (i.e., CPUs). The graphic processing unit (GPU) is one such accelerator. Originally designed solely for graphics processing, GPUs have evolved into programmable processors that can deliver massive parallel processing power for general-purpose applications. Using SIMD (Single Instruction Multiple Data) based components as building units; the current GPU architecture is well suited for data-parallel applications where the execution of each task is independent. With the delivery of programming models such as Compute Unified Device Architecture (CUDA) and Open Computing Language (OpenCL), programming GPUs has become much easier than before. However, developing and optimizing an application on a GPU is still a challenging task, even for well-trained computing experts. Such programming tasks will be even more challenging in large-scale heterogeneous systems, particularly in the context of utility computing, where GPU resources are used as a service. These challenges are largely due to the limitations in the current programming models: (1) there are no intra-and inter-GPU cooperative mechanisms that are natively supported; (2) current programming models only support the utilization of GPUs installed locally; and (3) to use GPUs on another node, application programs need to explicitly call application programming interface (API) functions for data communication. To reduce the mapping efforts and to better utilize the GPU resources, we investigate generalizing the utility of GPUs in large-scale heterogeneous systems with GPUs as accelerators. We generalize the utility of GPUs through the transparent virtualization of GPUs, which can enable applications to view all GPUs in the system as if they were installed locally. As a result, all GPUs in the system can be used as local GPUs. Moreover, GPU virtualization is a key capability to support the notion of "GPU as a service." Specifically, we propose the virtual OpenCL (or VOCL) framework for the transparent virtualization of GPUs. To achieve good performance, we optimize and extend the framework in three aspects: (1) optimize VOCL by reducing the data transfer overhead between the local node and remote node; (2) propose GPU synchronization to reduce the overhead of switching back and forth if multiple kernel launches are needed for data communication across different compute units on a GPU; and (3) extend VOCL to support live virtual GPU migration for quick system maintenance and load rebalancing across GPUs. With the above optimizations and extensions, we thoroughly evaluate VOCL along three dimensions: (1) show the performance improvement for each of our optimization strategies; (2) evaluate the overhead of using remote GPUs via several microbenchmark suites as well as a few real-world applications; and (3) demonstrate the overhead as well as the benefit of live virtual GPU migration. Our experimental results indicate that VOCL can generalize the utility of GPUs in large-scale systems at a reasonable virtualization and migration cost. / Ph. D.
3

Serverless Computing Strategies on Cloud Platforms

Naranjo Delgado, Diana María 08 February 2021 (has links)
[ES] Con el desarrollo de la Computación en la Nube, la entrega de recursos virtualizados a través de Internet ha crecido enormemente en los últimos años. Las Funciones como servicio (FaaS), uno de los modelos de servicio más nuevos dentro de la Computación en la Nube, permite el desarrollo e implementación de aplicaciones basadas en eventos que cubren servicios administrados en Nubes públicas y locales. Los proveedores públicos de Computación en la Nube adoptan el modelo FaaS dentro de su catálogo para proporcionar computación basada en eventos altamente escalable para las aplicaciones. Por un lado, los desarrolladores especializados en esta tecnología se centran en crear marcos de código abierto serverless para evitar el bloqueo con los proveedores de la Nube pública. A pesar del desarrollo logrado por la informática serverless, actualmente hay campos relacionados con el procesamiento de datos y la optimización del rendimiento en la ejecución en los que no se ha explorado todo el potencial. En esta tesis doctoral se definen tres estrategias de computación serverless que permiten evidenciar los beneficios de esta tecnología para el procesamiento de datos. Las estrategias implementadas permiten el análisis de datos con la integración de dispositivos de aceleración para la ejecución eficiente de aplicaciones científicas en plataformas cloud públicas y locales. En primer lugar, se desarrolló la plataforma CloudTrail-Tracker. CloudTrail-Tracker es una plataforma serverless de código abierto basada en eventos para el procesamiento de datos que puede escalar automáticamente hacia arriba y hacia abajo, con la capacidad de escalar a cero para minimizar los costos operativos. Seguidamente, se plantea la integración de GPUs en una plataforma serverless local impulsada por eventos para el procesamiento de datos escalables. La plataforma admite la ejecución de aplicaciones como funciones severless en respuesta a la carga de un archivo en un sistema de almacenamiento de ficheros, lo que permite la ejecución en paralelo de las aplicaciones según los recursos disponibles. Este procesamiento es administrado por un cluster Kubernetes elástico que crece y decrece automáticamente según las necesidades de procesamiento. Ciertos enfoques basados en tecnologías de virtualización de GPU como rCUDA y NVIDIA-Docker se evalúan para acelerar el tiempo de ejecución de las funciones. Finalmente, se implementa otra solución basada en el modelo serverless para ejecutar la fase de inferencia de modelos de aprendizaje automático previamente entrenados, en la plataforma de Amazon Web Services y en una plataforma privada con el framework OSCAR. El sistema crece elásticamente de acuerdo con la demanda y presenta una escalado a cero para minimizar los costes. Por otra parte, el front-end proporciona al usuario una experiencia simplificada en la obtención de la predicción de modelos de aprendizaje automático. Para demostrar las funcionalidades y ventajas de las soluciones propuestas durante esta tesis se recogen varios casos de estudio que abarcan diferentes campos del conocimiento como la analítica de aprendizaje y la Inteligencia Artificial. Esto demuestra que la gama de aplicaciones donde la computación serverless puede aportar grandes beneficios es muy amplia. Los resultados obtenidos avalan el uso del modelo serverless en la simplificación del diseño de arquitecturas para el uso intensivo de datos en aplicaciones complejas. / [CA] Amb el desenvolupament de la Computació en el Núvol, el lliurament de recursos virtualitzats a través d'Internet ha crescut granment en els últims anys. Les Funcions com a Servei (FaaS), un dels models de servei més nous dins de la Computació en el Núvol, permet el desenvolupament i implementació d'aplicacions basades en esdeveniments que cobreixen serveis administrats en Núvols públics i locals. Els proveïdors de computació en el Núvol públic adopten el model FaaS dins del seu catàleg per a proporcionar a les aplicacions computació altament escalable basada en esdeveniments. D'una banda, els desenvolupadors especialitzats en aquesta tecnologia se centren en crear marcs de codi obert serverless per a evitar el bloqueig amb els proveïdors del Núvol públic. Malgrat el desenvolupament alcançat per la informàtica serverless, actualment hi ha camps relacionats amb el processament de dades i l'optimització del rendiment d'execució en els quals no s'ha explorat tot el potencial. En aquesta tesi doctoral es defineixen tres estratègies informàtiques serverless que permeten demostrar els beneficis d'aquesta tecnologia per al processament de dades. Les estratègies implementades permeten l'anàlisi de dades amb a integració de dispositius accelerats per a l'execució eficient d'aplicacion scientífiques en plataformes de Núvol públiques i locals. En primer lloc, es va desenvolupar la plataforma CloudTrail-Tracker. CloudTrail-Tracker és una plataforma de codi obert basada en esdeveniments per al processament de dades serverless que pot escalar automáticament cap amunt i cap avall, amb la capacitat d'escalar a zero per a minimitzar els costos operatius. A continuació es planteja la integració de GPUs en una plataforma serverless local impulsada per esdeveniments per al processament de dades escalables. La plataforma admet l'execució d'aplicacions com funcions severless en resposta a la càrrega d'un arxiu en un sistema d'emmagatzemaments de fitxers, la qual cosa permet l'execució en paral·lel de les aplicacions segon sels recursos disponibles. Este processament és administrat per un cluster Kubernetes elàstic que creix i decreix automàticament segons les necessitats de processament. Certs enfocaments basats en tecnologies de virtualització de GPU com rCUDA i NVIDIA-Docker s'avaluen per a accelerar el temps d'execució de les funcions. Finalment s'implementa una altra solució basada en el model serverless per a executar la fase d'inferència de models d'aprenentatge automàtic prèviament entrenats en la plataforma de Amazon Web Services i en una plataforma privada amb el framework OSCAR. El sistema creix elàsticament d'acord amb la demanda i presenta una escalada a zero per a minimitzar els costos. D'altra banda el front-end proporciona a l'usuari una experiència simplificada en l'obtenció de la predicció de models d'aprenentatge automàtic. Per a demostrar les funcionalitats i avantatges de les solucions proposades durant esta tesi s'arrepleguen diversos casos d'estudi que comprenen diferents camps del coneixement com l'analítica d'aprenentatge i la Intel·ligència Artificial. Això demostra que la gamma d'aplicacions on la computació serverless pot aportar grans beneficis és molt àmplia. Els resultats obtinguts avalen l'ús del model serverless en la simplificació del disseny d'arquitectures per a l'ús intensiu de dades en aplicacions complexes. / [EN] With the development of Cloud Computing, the delivery of virtualized resources over the Internet has greatly grown in recent years. Functions as a Service (FaaS), one of the newest service models within Cloud Computing, allows the development and implementation of event-based applications that cover managed services in public and on-premises Clouds. Public Cloud Computing providers adopt the FaaS model within their catalog to provide event-driven highly-scalable computing for applications. On the one hand, developers specialized in this technology focus on creating open-source serverless frameworks to avoid the lock-in with public Cloud providers. Despite the development achieved by serverless computing, there are currently fields related to data processing and execution performance optimization where the full potential has not been explored. In this doctoral thesis three serverless computing strategies are defined that allow to demonstrate the benefits of this technology for data processing. The implemented strategies allow the analysis of data with the integration of accelerated devices for the efficient execution of scientific applications on public and on-premises Cloud platforms. Firstly, the CloudTrail-Tracker platform was developed to extract and process learning analytics in the Cloud. CloudTrail-Tracker is an event-driven open-source platform for serverless data processing that can automatically scale up and down, featuring the ability to scale to zero for minimizing the operational costs. Next, the integration of GPUs in an event-driven on-premises serverless platform for scalable data processing is discussed. The platform supports the execution of applications as serverless functions in response to the loading of a file in a file storage system, which allows the parallel execution of applications according to available resources. This processing is managed by an elastic Kubernetes cluster that automatically grows and shrinks according to the processing needs. Certain approaches based on GPU virtualization technologies such as rCUDA and NVIDIA-Docker are evaluated to speed up the execution time of the functions. Finally, another solution based on the serverless model is implemented to run the inference phase of previously trained machine learning models on theAmazon Web Services platform and in a private platform with the OSCAR framework. The system grows elastically according to demand and is scaled to zero to minimize costs. On the other hand, the front-end provides the user with a simplified experience in obtaining the prediction of machine learning models. To demonstrate the functionalities and advantages of the solutions proposed during this thesis, several case studies are collected covering different fields of knowledge such as learning analytics and Artificial Intelligence. This shows the wide range of applications where serverless computing can bring great benefits. The results obtained endorse the use of the serverless model in simplifying the design of architectures for the intensive data processing in complex applications. / Naranjo Delgado, DM. (2021). Serverless Computing Strategies on Cloud Platforms [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/160916

Page generated in 0.1076 seconds