Autodimensionamento proativo para elasticidade assertiva de funções de rede virtuais em cadeias de serviço

Abstract

The main idea of Network Function Virtualization is the decoupling of physic network equipment from the functions that run on then, which could then be implemented in software running on industrial standard physical servers. This flexibilization enables the deployment of network services provided over cloud-based infrastructures and the ability to enforces network policies through the chaining of different functions in which some traffic should traverse to accomplish it. The fluctuation of service demands will require from resource management the continuous elasticity of resources to meet cost and performance objectives, by allocating or releasing resources to avoid under/over-provisioning. Traditionally, elasticity is offered through reactive approaches based on threshold-based rules to perform elastic actions, although they tend to the scarcity of resources and increasing service response time. Recent work has suggested proactive approaches, through combining Machine Learning-based models which allows anticipating decisions and optimize the adjustment of resources to demand. Such optimization is crucial for a successful proactive elasticity solution, which enables assertive auto-scaling decisions to rapidly respond to demand fluctuation and contributes to the balance of cost-performance objectives. This doctoral thesis presents ARENA, a proactive autoscaling mechanism with Machine Learning-based demand prediction for maximization of assertive decisions in horizontal and vertical elasticity.