Cloud federation formation using genetic and evolutionary game theoretical models

This paper proposes an approach based on genetic algorithms and evolutionary game theory in order to study the problem of forming highly profitable federated clouds, while maintaining stability among the members in the presence of dynamic strategies (i.e. cloud providers joining and/or leaving federations) that might result in decreased Quality of Service (QoS). Cloud federation helps cloud providers to take advantage of the available unused virtual machines. It allows the providers to combine their resources in order to serve a larger pool of requests that could not have been served otherwise. We tackle the problem of forming federations while maximizing the total profit they yield using a Genetic Algorithm. However, the main problem may rise after the federation formation where many cloud providers, due to the dynamicity, may be tempted to reallocated their resources into other federations for seeking better payoff. Such an act may lead to a decrease in the QoS and cause a drop in the profit earned by the federations. Thus, we extend the genetic model as an evolutionary game, which aims to improve the profit while maintaining stability among federations. Experiments were conducted using CloudHarmony real-world dataset and benchmarked with Sky federation model previously introduced in the literature. Both the genetic and evolutionary game theoretical models outperform the benchmarked one. The evolutionary game model gave better results in terms of profit and QoS’s due to its mechanism of reaching a stable state, in which no provider has incentive to reallocate his resources into different federations.