CARVALHO, M W A; http://lattes.cnpq.br/4764601053478564; CARVALHO, Marcus Williams Aquino de.
Resumen:
Infrastructure as a Service (IaaS) is a cloud computing model that has been growing significantly in recent years. This increasingly adoption attracted users with different types of applications, requirements and budget to the cloud. To satisfy different users’s needs, IaaS providers can offer multiple service classes with different pricing and Service Level Objectives (SLOs) defined for them. However, managing such clouds considering multiple service classes is nottrivial, becauseres our cemanagement decisions may have different consequences depending on howeach classisaffected. Moreover, the demand elasticity and uncertain resource availability typically seen in cloudenvironments turns evenmore difficult for providers to fulfill different SLOs while having a high utilization and low infrastructure costs. In this thesis, we investigate the hypothesis that when IaaS cloud providers make an adequate resource management, offering multiple service classes and fulfilling their Quality of Service (QoS) guarantees, they achieve a high resource utilization and increase their revenue. We demonstrate that this hypothesis is true for the many different scenarios evaluated in this work,which shows great advantages on offering multiple service classes in the cloud. However, we also observe that cloud providers need an efficient resource management in order to have these benefits, in a way they can define and fulfill adequate SLOs for different classes. Thus, we also show in this thesis how IaaS providers can make adequate resource management decisions for multiple service classes, by proposing and evaluating: (1) a predictive method to planthe capacityand QoS guarantees of anew service class introducedin an existing IaaS cloud, based on unused resources; (2) a prediction-based admission control model that allows the provider to offer multiple classes and fulfill VM availability SLOs for them; and (3) a capacity planning analytical model that estimates QoS metrics for eachclass indifferents cenarios, and aimstofindthem in imumres ource capacity required to fulfill VM availability and admission rate SLOs for eachclass.