Alocação de Ambientes Virtuais com base na Afinidade entre Perfis de Aplicações Massivamente Paralelas e Distribuída

Abstract

Afinidade" entre aplicações. Para a implementação do método, foi realizado o monitoramento do consumo de recursos das máquinas virtuais para obter dados históricos de execução que permitem determinar os perfis das aplicações. Como resultado, foi criado um escalonador para alocar máquinas virtuais, cujos objetivos são evitar combinações de aplicações que causem degradação do desempenho entre si e melhorar a taxa de utilização dos recursos computacionais.

Application of Profile Prediction for Proactive Scheduling

“…Oliveira et al [5] presents a scheduler capable of identifying low affinity between applications running on the same PM and migrating one of the applications to another machine, freeing resources and interrupting interference. The system uses empirically defined consumption thresholds and an Exponential Moving Average (EMA) to identify trends in consumption that can detect changes in application profiles at runtime.…”

“…The architecture proposed by Oliveira et al [5] consists of several monitors (one for each application) and a central scheduler. These monitors collect application resource usage data while assessing changes in consumption profile, notifying the scheduler of detected changes.…”

“…Our approach consists of using machine learning algorithms in order to predict the computational resources con-sumption profile of applications based on their monitoring, thus, acting proactively. We chose the architecture proposed by Oliveira et al [5] to be the basis of our work because it allows the addition of new strategies for monitoring and identifying profile changes. Figure 1 illustrates the general architecture of the scheduling system with a central scheduler and multiple pairs of VMs and monitors.…”

“…Yokoyama et al [4] proposed, for example, the estimation of the affinity of applications and then uses this static value to schedule these applications on the most appropriate hosts to avoid interference. Some works also monitor the VM resources consumption of each application to identify profile changes and reallocate VMs through migrations [5].…”

“…Affinity-based scheduling models [4] show good results by avoiding running applications with low affinity on the same real machine, for those that always maintain the same profile. Furthermore, with real-time monitoring of virtual machines, it is possible to detect when an application changes its profile and starts to cause, or suffer from, interference [5], allowing a corrective reaction such as the migration of the VM, for instance.…”