Arquiteturas reconfiguráveis têm se demonstrado uma potencial solução para lidar com a crescente complexidade encontrada em sistemas embarcados. Para se alcançar ganhos em desempenho, é preciso uma grande redundância das unidades funcionais, acarretando o aumento da área ocupada pelas unidades funcionais. Uma das propostas deste trabalho será de explorar o espaço de projeto, visando à redução da área e da energia. Para isto, serão apresentadas duas técnicas de virtualização de hardware, sendo as mesmas semelhantes a um pipeline de estágios reconfiguráveis. Ambas as técnicas alcançaram mais de 94% de redução da área. Outro aspecto a ser explorado em uma arquitetura reconfigurável é o impacto em área e energia causado pela inserção da memória de contexto. Assim, este impacto será demonstrado neste trabalho e duas abordagens que modificam a memória de contexto serão propostas: a primeira abordagem baseia-se na exploração da largura ideal da porta da memória combinado com número de acessos, para que se minimize a energia consumida na busca dos bytes de configuração; a segunda abordagem possui um mecanismo de gerenciamento das configurações por meio de listas ligadas, que permite que as configurações sejam acessadas parcialmente. As duas abordagens apresentaram redução de energia de até 98%, podendo ser utilizadas em sistemas que apresentam tanto a reconfiguração parcial como a total. / Reconfigurable architectures have shown to be a potential solution to the problem of increasing complexity found in embedded systems. However, in order to achieve significant performance gains, large quantities of redundant functional units are generally necessary, with a corresponding increase in the area occupied by these units. This thesis explores the design space with the objective of reducing both area and energy consumption, and presents two hardware virtualization techniques, similar to reconfigurable pipeline stages, which achieve a reduction in area of more than 94%. The use of context memory in reconfigurable architectures has a significant impact in terms of area and energy, as is clearly demonstrated by initial experimental results. Two novel context memory architectures are presented: the first approach is being based on an exploration of the balance point between memory port width and number of accesses, in order to reduce the energy consumed during fetching of the configuration bytes; the second approach presents a configuration management mechanism using hardware linked lists, and that allows segmented access to configuration settings. Both approaches demonstrate energy reduction of up to 98% and can be adopted in both partial and atomic reconfiguration architectures.

Arquiteturas reconfiguráveis têm se demonstrado uma potencial solução para lidar com a crescente complexidade encontrada em sistemas embarcados. Para se alcançar ganhos em desempenho, é preciso uma grande redundância das unidades funcionais, acarretando o aumento da área ocupada pelas unidades funcionais. Uma das propostas deste trabalho será de explorar o espaço de projeto, visando à redução da área e da energia. Para isto, serão apresentadas duas técnicas de virtualização de hardware, sendo as mesmas semelhantes a um pipeline de estágios reconfiguráveis. Ambas as técnicas alcançaram mais de 94% de redução da área. Outro aspecto a ser explorado em uma arquitetura reconfigurável é o impacto em área e energia causado pela inserção da memória de contexto. Assim, este impacto será demonstrado neste trabalho e duas abordagens que modificam a memória de contexto serão propostas: a primeira abordagem baseia-se na exploração da largura ideal da porta da memória combinado com número de acessos, para que se minimize a energia consumida na busca dos bytes de configuração; a segunda abordagem possui um mecanismo de gerenciamento das configurações por meio de listas ligadas, que permite que as configurações sejam acessadas parcialmente. As duas abordagens apresentaram redução de energia de até 98%, podendo ser utilizadas em sistemas que apresentam tanto a reconfiguração parcial como a total. / Reconfigurable architectures have shown to be a potential solution to the problem of increasing complexity found in embedded systems. However, in order to achieve significant performance gains, large quantities of redundant functional units are generally necessary, with a corresponding increase in the area occupied by these units. This thesis explores the design space with the objective of reducing both area and energy consumption, and presents two hardware virtualization techniques, similar to reconfigurable pipeline stages, which achieve a reduction in area of more than 94%. The use of context memory in reconfigurable architectures has a significant impact in terms of area and energy, as is clearly demonstrated by initial experimental results. Two novel context memory architectures are presented: the first approach is being based on an exploration of the balance point between memory port width and number of accesses, in order to reduce the energy consumed during fetching of the configuration bytes; the second approach presents a configuration management mechanism using hardware linked lists, and that allows segmented access to configuration settings. Both approaches demonstrate energy reduction of up to 98% and can be adopted in both partial and atomic reconfiguration architectures.

Arquiteturas reconfiguráveis têm se demonstrado uma potencial solução para lidar com a crescente complexidade encontrada em sistemas embarcados. Para se alcançar ganhos em desempenho, é preciso uma grande redundância das unidades funcionais, acarretando o aumento da área ocupada pelas unidades funcionais. Uma das propostas deste trabalho será de explorar o espaço de projeto, visando à redução da área e da energia. Para isto, serão apresentadas duas técnicas de virtualização de hardware, sendo as mesmas semelhantes a um pipeline de estágios reconfiguráveis. Ambas as técnicas alcançaram mais de 94% de redução da área. Outro aspecto a ser explorado em uma arquitetura reconfigurável é o impacto em área e energia causado pela inserção da memória de contexto. Assim, este impacto será demonstrado neste trabalho e duas abordagens que modificam a memória de contexto serão propostas: a primeira abordagem baseia-se na exploração da largura ideal da porta da memória combinado com número de acessos, para que se minimize a energia consumida na busca dos bytes de configuração; a segunda abordagem possui um mecanismo de gerenciamento das configurações por meio de listas ligadas, que permite que as configurações sejam acessadas parcialmente. As duas abordagens apresentaram redução de energia de até 98%, podendo ser utilizadas em sistemas que apresentam tanto a reconfiguração parcial como a total. / Reconfigurable architectures have shown to be a potential solution to the problem of increasing complexity found in embedded systems. However, in order to achieve significant performance gains, large quantities of redundant functional units are generally necessary, with a corresponding increase in the area occupied by these units. This thesis explores the design space with the objective of reducing both area and energy consumption, and presents two hardware virtualization techniques, similar to reconfigurable pipeline stages, which achieve a reduction in area of more than 94%. The use of context memory in reconfigurable architectures has a significant impact in terms of area and energy, as is clearly demonstrated by initial experimental results. Two novel context memory architectures are presented: the first approach is being based on an exploration of the balance point between memory port width and number of accesses, in order to reduce the energy consumed during fetching of the configuration bytes; the second approach presents a configuration management mechanism using hardware linked lists, and that allows segmented access to configuration settings. Both approaches demonstrate energy reduction of up to 98% and can be adopted in both partial and atomic reconfiguration architectures.

Cette thèse s'intéresse aux architectures contenant des FPGAs reconfigurables dynamiquement et partiellement. Dans ces architectures, la complexité et la difficulté de portage des applications sont principalement dues aux connections étroites entre la gestion de la reconfiguration et le calcul lui-même. Nous proposons 1) un nouveau niveau d'abstraction, appelé gestionnaire de composants matériels (HCM) et 2) un mécanisme de communication scalable (SCM), qui permettent une séparation claire entre l'allocation d'une fonction matérielle et la procédure de reconfiguration. Cela réduit l'impact de la gestion de la reconfiguration dynamique sur le code de l'application, ce qui simplifie grandement l'utilisation des plateformes FPGA. Les application utilisant le HCM et le SCM peuvent aussi être portées de manière transparentes à des systèmes multi-FPGA et/ou multi-utilisateurs. L'implémentation de cette couche HCM et du mécanisme SCM sur des plateformes réalistes de prototypage virtuel démontre leur capacité à faciliter la gestion du FPGA tout en préservant les performances d'une gestion manuelle, et en garantissant la protection des fonctions matérielles. L'implémentation du HCM et du mécanisme SCM ainsi que leur environnement de simulation sont open-source dans l'espoir d'une réutilisation par la communauté. / This thesis shows that in FPGA-based dynamic reconfigurable architectures, the complexity and low portability of application developments are mainly due to the tight connections between reconfiguration management and computation. By proposing 1) a new abstraction layer, called Hardware Component Manager (HCM) and 2) a Scalable Communication Mechanism (SCM), we clearly separate the allocation of a hardware function from the control of a reconfiguration procedure. This reduces the dynamic reconfiguration management impact on the application code, which greatly simplifies the use of FPGA platforms. Applications using the HCM and the SCM can also be transparently ported to multi-user and/or multi-FPGA systems. The implementation of this HCM layer and the SCM mechanism on realistic simulation platforms demonstrates their ability to ease the management of FPGA flexibility while preserving performance and ensuring hardware function protection. The HCM and SCM implementations and their simulation environment are open-source in the hope of reuse by the community.

De part leur capacité de reconfiguration et les performances qu’ils offrent, les FPGAs sont de bons candidats pour accélérer des applications dans le Cloud. Cependant, les FPGAs présentent certaines caractéristiques qui font obstacle à leur utilisation dans le Cloud et leur adoption par les clients : premièrement, la programmation des FPGAs se fait à bas niveau et demande une certaine expertise, que n’ont pas nécessairement les clients habituels du Cloud. Deuxièmement, les FPGAs ne présentent pas de mécanismes natifs permettant leur intégration dans le modèle de gestion dynamique d’une infrastructure Cloud.Dans ce travail, nous proposons d’utiliser des architectures overlay afin de faciliter l’adoption, l’intégration et l’exploitation de FPGAs dans le Cloud. Les overlays sont des architectures reconfigurables elles-mêmes implémentée sur FPGA. En tant que couche d’abstraction matérielle placée entre le FPGA et les applications, les overlays permettent de monter le niveau d’abstraction du modèle d’exécution présenté aux applications et aux utilisateurs, ainsi que d’implémenter des mécanismes facilitant leur intégration et leur exploitation dans une infrastructure Cloud.Ce travail présente une approche verticale adressant tous les aspects de la mise en œuvre d’overlays dans le Cloud en tant qu’accélérateurs reconfigurables par les clients : de la conception et l’implémentation des overlays, leur intégration sur des plateformes FPGA commerciales, la mise en place de leurs mécanismes d’exploitation, jusqu’à la réalisationde leurs outils de programmation. L’environnement réalisé est complet, modulaire et extensible, il repose en partie sur différents outils existants, et démontre la faisabilité de notre approche. / Due to their reconfigurable capability and the performance they offer, FPGAs are good candidates for accelerating applications in the cloud. However, FPGAs have some features that hinder their use in the Cloud as well as their adoption by customers : first, FPGA programming is done at low level and requires some expertise that usual Cloud clients do not necessarily have. Secondly, FPGAs do not have native mechanisms allowing them to easily fit in the dynamic execution model of the Cloud.In this work, we propose to use overlay architectures to facilitate FPGA adoption, integration, and operation in the Cloud. Overlays are reconfigurable architectures synthesized on FPGA. As hardware abstraction layers placed between the FPGA and applications, overlays allow to raise the abstraction level of the execution model presented to applications and users, as well as to implement mechanisms making them fit in a Cloud infrastructure.This work presents a vertical approach addressing all aspects of overlay operation in the Cloud as reconfigurable accelerators programmable by tenants : from designing and implementing overlays, integrating them on commercial FPGA platforms, setting up their operating mechanisms, to developping their programming tools. The environment developped in this work is complete, modular and extensible, it is partially based on several existing tools, and demonstrate the feasibility of our approach.

Virtualization is a technique used to abstract the operating system from the hardware. The primary gains of virtualization is increased server consolidation, leading to greater hardware utilization and infrastructure manageability. Another technology that can be used to achieve similar goals is containerization. Containerization is an operating-system level virtualization technique which allows applications to run in partial isolation on the same hardware. Containerized applications share the same Linux kernel but run in packaged containers which includes just enough binaries and libraries for the application to function. In recent years it has become more common to see hardware virtualization beneath the container host operating systems. An upcoming technology to further this development is VMware’s vSphere Integrated Containers which aims to integrate management of Linux Containers with the vSphere (a hardware virtualization platform by VMware) management interface. With these technologies as background we set out to measure the impact of hardware virtualization on Linux Container performance by running a suite of macro-benchmarks on a LAMP-application stack. We perform the macro-benchmarks on three different operating systems (CentOS, CoreOS and Photon OS) in order to see if the choice of container host affects the performance. Our results show a decrease in performance when comparing a hardware virtualized container host to a container hosts running directly on the hardware. However, the impact on containerized application performance can vary depending on the actual application, the choice of operating system and even the type of operation performed. It is therefore important to consider these three items before implementing container hosts as virtual machines.