Cost minimization of service deployment in a multi-cloud environment

Abstract

Public cloud computing allows one to rent virtual servers on a hourly basis. This raises the problematic of being able to decide which server offers to take, which providers to use, and how to use them to acquire sufficient service capacity, while maintaining a cost effective platform. This article proposes a new realistic model to tackle the problem, placing services into IAAS virtual machines from multiple providers. A flexible protocol is defined to generate real-life instances, and applied on two industrial cases with four real cloud providers. An evolutionary approach, with new specific operators, is introduced and compared to a MIP formulation. Experiments conducted on two data-sets show that the evolutionary approach is viable to tackle real-size instances in reasonable amount of time.