SANTOS, M. T. G.; http://lattes.cnpq.br/1067342907017432; SANTOS, Marcela Tassyany Galdino.
Resumo:
The Infrastructure as a Service (IaaS), it has attracted more user, which host the most several application with differents requirements. That it happens because of the cost reduction, management facility and scalability. Usually, the clients tend to overestimate their resources necessity, that can cause subutilization, waste of energy and, consequently, more cost to providers. Though, the providers have been search benefit with this scenario when admite more VMs than they can support. This technique is called by Overbooking. However, maximizing resources use can result on workloads performance reduction in execution. In that regard, has a trade-off between grow up resources utilization and the threat of performance degradation. In this work, is proposed a approach to overbooking management, it aims to propose an overbooking level that promote a better commitment between provider and client. Our approach use the Analytic Hierarchy Process (AHP) to obtain user’s indication related to service type, based on screening algorithm used in hospital in whole world, to propose a classification of VMs according to theirs criticality. Furthermore, it was developed a approach to dynamic
attribution of overbooking based on linear regression technique, which from monitoring data is created a CPU prevision model for each VM. Thus, is possible define a overbooking level what better suits to VM utilization profile and its criticality class. The results, compared to two other overbooking approach, demonstrate that proposed approach has a better balance in there trade-off, being possible to take overbooking benefits and reduct the performance impact in the application.