Gatekeeper: controle de tráfego distribuído em datacenters virtualizados

• Ano de defesa: 2010
• Autor(a) principal: Paolo Victor Goncalves Soares
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; UFMG
• Palavras-chave em Português: Cloud computing; Virtualização; Controle de tráfego de rede; Sistemas distribuídos; Computação; Computação em nuvem; Redes de computadores
• Link de acesso: http://hdl.handle.net/1843/BUBD-9W7FAR

A datacenter hosts multiple services that share a single physical infrastructure. On these environments, the sharing of computational resources among the different services is implemented by virtualization technologies, that create collections of virtual machines for each service. In the Cloud Computing model, its expected that services that belong to different users end up sharing the same physical resources, including processors, memory and network access. This urges the implementation of proper resource sharing and isolation between these virtual machines, for both security and efficiency reasons. Even though the current virtualization technologies support the isolation of processing and memory resources, the current network sharing technologies, that control the transmission and reception rates of the virtual machines, are only reliable if the virtual machines traffic implements some form of flow control algorithm, like TCP traffic, or if certain hardware elements are modified or added to the network trunk. The consolidation of high-redundancy, high-performance network architectures that can be built with simpler hardware suggests that a server-side solution that overcomes the current technologies limitations is needed. This work presents Gatekeeper, a distributed traffic control mechanism that operates on the virtualization layer. Gatekeeper is transparent to the virtual machines hosted at the datacenter, and offers both transmission and reception rate guarantees for each virtual machine, the later being ensured by a distributed network congestion detection and mitigation algorithm.