Title: Learning techniques for cloud demands aggregation in cloud service brokering
Authors: Dominique Barth; Christian Cadéré; Johanne Cohen; Thierry Mautor; Sandrine Vial
Addresses: DAVID, UVSQ, Université Paris-Saclay, 45 Avenue des Etats-Unis, 78035 VERSAILLES Cedex, France ' DAVID, UVSQ, Université Paris-Saclay, 45 Avenue des Etats-Unis, 78035 VERSAILLES Cedex, France ' LRI, Univ. Paris-Sud, CNRS, Université Paris-Saclay, Bat 650, 91405 Orsay Cedex, France ' DAVID, UVSQ, Université Paris-Saclay, 45 Avenue des Etats-Unis, 78035 VERSAILLES Cedex, France ' DAVID, UVSQ, Université Paris-Saclay, 45 Avenue des Etats-Unis, 78035 VERSAILLES Cedex, France
Abstract: The cloud service broker (CSB) has to ensure intermediation for cloud services. Especially, it acts as an interface among customers and providers of cloud resources. The main purpose of the CSB is to pertinently aggregate different demands of customers into a global demand. The objective is to obtain a price for this global demand as low as possible from the providers. Finding the best aggregations of demands is an optimisation problem. In a competitive context between providers, this optimisation problem has to be solved taking into account partial knowledge. We model this problem as a game in which we evaluate the quality of Nash equilibria from the optimisation problem point of view. We propose to use a linear reward inaction algorithm to reach such equilibria in a partial knowledge and distributed context after proving a convergence property of the algorithm. This approach is finally experimented by numerous simulations.
Keywords: game theory; cloud computing; optimisation; learning techniques; cloud services; cloud service brokering; cloud computing; Nash equilibria; simulation; partial knowledge.
International Journal of Cloud Computing, 2016 Vol.5 No.4, pp.283 - 308
Accepted: 15 Nov 2015
Published online: 11 Dec 2016 *