MSIALLOCATOR: aplicação da meta-heurística ILS para o problema de alocação de máquinas virtuais em uma Data Center

Abstract

Cloud Computing is a new computing model where customers outsource computing resources, such as the ability to process and store data, consuming them as a service over the Internet. In this paradigm, the resources are served through a provider that has a large physical infrastructure of server machines, this infrastructure is known as a data center. Customers consume such resources through virtual machines that are distributed among the numerous server machines that make up the data center. The increasing use of this paradigm has caused a rapid expansion in the infrastructure of the providers, which causes them to consume a lot of electric energy. Because of this, several virtual machine allocation policies have been proposed with the goal of minimizing the energy costs of a data center through the intelligent allocation of virtual machines to physical servers. It is understood as intelligent allocation the action of distributing the virtual machines between the physical servers in such a way that it is possible to generate energy savings in the data center, but without degrading the performance of the services provided. This work presents a new allocation policy, called MSIAllocator, which addresses the Virtual Machine Allocation Problem in a similar way to the Multiple Backpack Problem, using the Iterated Local Search Meta-Heuristic (ILS), which generates a random solution the current solution through perturbations and local searches. This new policy saved more electricity than all the other policies analyzed in this study, consuming an average of 109 kWh per day, representing a saving of 28% if compared to the less economic policy. It also left the number of migrations and SLA violations at acceptable levels, with 21486 and 5.99% per day, respectively.