| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1.7 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
A virtualização nativa ao nível do hardware é atualmente usada em datacenters para suporte
à virtualização de infraestruturas de computação de elevada disponibilidade. Os hypervisors
nativos têm sido a principal escolha para a construção de infraestruturas virtualizadas baseadas
em failover clusters usando o Failover Clustering no caso do Microsoft Windows Hyper-V.
Contudo, o recente surgimento da tecnologia Docker, a tecnologia de containers de software
mais usada a nível mundial, tem tornado popular a virtualização ao nível do sistema operativo,
permitindo criar, através do modo Swarm, infraestruturas virtualizadas de elevada disponibilidade.
A virtualização ao nível do hardware apresenta vantagens em relação à virtualização ao
nível do sistema operativo, de entre as quais se destacam o isolamento e a flexibilidade de aplicações,
mas apresenta desvantagens tais como uma maior complexidade de implementação e
lentidão na inicialização e no armazenamento de imagens de máquinas virtuais. Por outro lado,
a virtualização nativa ao nível do hardware tem evoluído no sentido de melhorar o respetivo
desempenho, em que as arquiteturas de hypervisors baseadas em virtualização completa com
tradução binária estão a ser substituídas por arquiteturas de hypervisors com paravirtualização
com suporte por hardware. Esta dissertação pretende avaliar e comparar o desempenho de duas
infraestruturas virtualizadas de elevada disponibilidade, uma baseada na virtualização nativa ao
nível do hardware usando o Microsoft Hyper-V com Failover Clustering e a outra baseada na virtualização
ao nível do sistema operativo usando o Docker em modo Swarm. As infraestruturas
virtualizadas implementadas consistiram num failover cluster com dois nós físicos e um servidor
de armazenamento. Foi avaliado e comparado o desempenho destas infraestruturas usando quatro
diferentes benchmarks, o IOzone cujo alvo de avaliação é o sistema de ficheiros, o RAMspeed
que permite a medição do desempenho da memória do sistema, o iPerf3 para medições ativas
da largura de banda na comunicação entre plataformas e o Geekbench que é um benchmark de
processadores multiplataforma. Os resultados obtidos com estes quatro benchmarks mostram,
de forma geral, que o desempenho do Docker em modo Swarm é superior ao desempenho do
Microsoft Hyper-V com Failover Clustering. Foram também analisados os tempos de migração
de máquinas virtuais e de containers naquelas duas infraestruturas virtualizadas, tendo-se observado
tempos de migração da ordem de 18,4s para máquinas virtuais com sistema operativo
Windows sobre o Hyper-V, tempos de migração da ordem de 15,35s para máquinas virtuais com
sistema operativo Linux sobre o Hyper-V e tempos de migração de Docker containers em modo
Swarm na ordem de 5,94s e tempos de migração de Nested Docker containers na ordem de
6,79s. Estes resultados mostram que o Docker apresenta um desempenho superior em relação
ao Hyper-V em termos de tempos de migração de máquinas virtuais ou de containers de um nó
para o outro do failover cluster.
Native virtualization at the hardware level is currently used in datacenters to support virtualization of high availability computing infrastructures. Native hypervisors have been the main choice for building virtualized infrastructures based on failover clusters using Failover Clustering in the case of Microsoft Windows Hyper-V. However, the recent emergence of Docker technology, the most widely used software container technology worldwide, has made virtualization popular at the operating system level, allowing the creation, through the Swarm mode, of virtualized infrastructures of high availability. Virtualization at the hardware level has advantages over virtualization at the operating system level, among which application isolation and flexibility stand out, but it has drawbacks such as greater implementation complexity and slow boot and storage of images of virtual machines. On the other hand, native hardware virtualization has evolved to improve performance, where hypervisor architectures based on full binary-based virtualization are being replaced by hardware-supported paravirtualization hypervisor architectures. This dissertation aims to evaluate and compare the performance of two highly available virtualized infrastructures, one based on native virtualization at the hardware level using Microsoft Hyper-V with Failover Clustering and the other based on operating system virtualization using Docker in Swarm mode. The virtualized infrastructures implemented consisted of a failover cluster with two physical nodes and a storage server. The performance of these infrastructures was evaluated and compared using four different benchmarks, the IOzone whose evaluation target is the file system, the RAMspeed that allows the measurement of system memory performance, iPerf3 for active measurements of bandwidth in communication between platforms and the Geekbench which is a benchmark of multiplatform processors. The results obtained with these four benchmarks generally show that Docker performance in Swarm mode is higher than Microsoft Hyper-V performance with Failover Clustering. The migration times of virtual machines and containers in these two virtualized infrastructures were also analyzed, with migration times of the order of 18.4s for virtual machines with Windows operating system on Hyper-V, migration times of the order of 15.35s for virtual machines with Linux operating system over Hyper-V and Docker containers migration times in Swarm mode in the order of 5.94s and migration times of Nested Docker containers in the order of 6.79s. These results show that Docker delivers superior performance over Hyper-V in terms of migration times from virtual machines or containers from one node to another from the failover cluster.
Native virtualization at the hardware level is currently used in datacenters to support virtualization of high availability computing infrastructures. Native hypervisors have been the main choice for building virtualized infrastructures based on failover clusters using Failover Clustering in the case of Microsoft Windows Hyper-V. However, the recent emergence of Docker technology, the most widely used software container technology worldwide, has made virtualization popular at the operating system level, allowing the creation, through the Swarm mode, of virtualized infrastructures of high availability. Virtualization at the hardware level has advantages over virtualization at the operating system level, among which application isolation and flexibility stand out, but it has drawbacks such as greater implementation complexity and slow boot and storage of images of virtual machines. On the other hand, native hardware virtualization has evolved to improve performance, where hypervisor architectures based on full binary-based virtualization are being replaced by hardware-supported paravirtualization hypervisor architectures. This dissertation aims to evaluate and compare the performance of two highly available virtualized infrastructures, one based on native virtualization at the hardware level using Microsoft Hyper-V with Failover Clustering and the other based on operating system virtualization using Docker in Swarm mode. The virtualized infrastructures implemented consisted of a failover cluster with two physical nodes and a storage server. The performance of these infrastructures was evaluated and compared using four different benchmarks, the IOzone whose evaluation target is the file system, the RAMspeed that allows the measurement of system memory performance, iPerf3 for active measurements of bandwidth in communication between platforms and the Geekbench which is a benchmark of multiplatform processors. The results obtained with these four benchmarks generally show that Docker performance in Swarm mode is higher than Microsoft Hyper-V performance with Failover Clustering. The migration times of virtual machines and containers in these two virtualized infrastructures were also analyzed, with migration times of the order of 18.4s for virtual machines with Windows operating system on Hyper-V, migration times of the order of 15.35s for virtual machines with Linux operating system over Hyper-V and Docker containers migration times in Swarm mode in the order of 5.94s and migration times of Nested Docker containers in the order of 6.79s. These results show that Docker delivers superior performance over Hyper-V in terms of migration times from virtual machines or containers from one node to another from the failover cluster.
Description
Keywords
Container Desempenho e Benchmarking. Docker Hyper-V Hypervisor Máquina Virtual Virtualização