Optimizing PaaS provider profit under service level agreement constraints. (Optimisation du profit des fournisseurs PaaS sous des contraintes de contrat de service)

Cloud computing is an emerging paradigm revolutionizing the use and marketing of informationtechnology. As the number of cloud users and providers grows, the socio-economical impact ofcloud solutions and particularly PaaS (platform as a service) solutions is becoming increasinglycritical. The main objective of PaaS providers is to generate the maximum profit from the servicesthey provide. This requires them to face a number of challenges such as efficiently managing theunderlying resources and satisfying the SLAs of the hosted applications.This thesis considers a cloud-bursting PaaS environment where the PaaS provider owns alimited number of private resources and is able to rent public cloud resources, when needed.This environment enables the PaaS provider to have full control over services hosted on theprivate cloud and to take advantage of public clouds for managing peak periods. In this context,we propose a profit-efficient solution for managing the cloud-bursting PaaS system under SLAconstraints. We define a profit optimization policy that, after each client request, evaluates thecost of hosting the application using public and private resources and chooses the option thatgenerates the highest profit. During peak periods the optimization policy considers two moreoptions. The first option is to take some resources from running applications, taking into accountthe payment of penalties if their promised quality of service is affected. The second option isto wait until private resources become available, taking into account the payment of penalties ifthe promised quality of service of the new application is affected. Furthermore we designed andimplemented an open cloud-bursting PaaS system, called Meryn, which integrates the proposedoptimization policy and provides support for batch and MapReduce applications. The results ofour evaluation show the effectiveness of our approach in optimizing the provider profit. Indeed,compared to a basic approach, our approach provides up to 11.59% and 9.02% more providerprofit in, respectively, simulations and experiments.