Cost Minimization of Virtual Machine Allocation in Public Clouds Considering Multiple Applications

Abstract

This paper presents a virtual machine (VM) allocation strategy to optimize the cost of VM deployments in public clouds. It can simultaneously deal with multiple applications and it is formulated as an optimization problem that takes the level of performance to be reached by a set of applications as inputs. It considers real characteristics of infrastructure providers such as VM types, limits on the number VMs that can be deployed, and pricing schemes. As output, it generates a VM allocation to support the performance requirements of all the applications. The strategy combines short-term and long-term allocation phases in order to take advantage of VMs belonging to two different pricing categories: on-demand and reserved. A quantization technique is introduced to reduce the size of the allocation problem and, thus, significantly decrease the computational complexity. The experiments show that the strategy can optimize costs for problems that could not be solved with previous approaches