Mapeamento de máquinas virtuais em datacenters privados visando minimizar a interferência de desempenho

Abstract

The power of computers increases year after year and today is common to have at home personal computers with computational power similar to servers and mainframes of years ago. Naturally, todays servers followed this evolution. However, many applications that run in these servers no longer require the computational power of a single, exclusive, server and this led to the development of solutions to avoid wasting servers resources. A common approach is server consolidation. Through virtualization it is possible to share resources from a single server among multiple virtual machines, reducing the waste of resources and increasing the amount of customers that can be served with a single server. However, several studies [19, 24, 38] show that virtual machines can interfer in the performance of other virtual machines when there are disputes over the same resources. This work proposes an algorithm for mapping virtual machines that minimize the number of servers required while maintaining the performance interference below a threshold specified by the user. The results obtained after evaluating the proposed solution show that it can map virtual machines without exceeding the threshold set by the user, as well as significantly reduce the interference without an expressive increase in the number of required servers.