Game theoretical models for cloud federations. (2019)

Abstract

Cloud federation is an architecture that allows cloud providers to make use of their unallocated virtual machines, by combining their resources to serve a pool of cloud consumers whose requests cannot be handled by any of these providers alone. A client is willing to rent computing resources from the cloud providers or federations due to the advantages they can provide. The quality of service (QoS) is one of the main factors that attracts or discourages the clients from renting such services. What complicates the process is having the actual QoS delivered being worse than the promised QoS. Such could lead to the client changing federation, and the latter getting dissociated. Some of the main reasons that could result in worsening the QoS are encountering passive malicious cloud providers after the federation formation, and having unstable federation formation. In this thesis, we present solutions for such problems in order to increase the lifespan of the formed federations, by introducing a maximin game to prevent the malicious providers from accomplishing their wicked schemes without getting penalized, and advancing a genetic and an evolutionary game theoretical models for the federation formation process to bypass the dynamicity boundaries. Experiments conducted using CloudHarmony real-world dataset revealed that both of our solutions were able to increase the total profit obtained by the federations and ameliorate the QoS delivered, granting the cloud consumer a great experience with the service.