An influence diagram based cloud service selection approach in dynamic cloud marketplaces

Abstract

Cloud service selection is one of the key issues for service users. An important concern for this issue is to make an optimal policy that adapt to a dynamic cloud marketplace that different types of changes may happen during a service consumption process, for example, the change of the cloud service providers and the provisioned cloud services in this marketplace, the change of the requirements of service users on service performance, e.g. Quality of services (QoSs) and service functions, and the change of the termination time points of a consumed service. These changes require a dynamic modelling method for cloud service selection, while existing service selection approaches rarely consider such dynamicity. Therefore, we propose a novel Cloud service selection framework based on Markov decision processes (MDPs), which can help Cloud service users to select a set of services meeting the QoS requirements and the economic constraints of service users. The MDP, as one of the primary decision theoretic planning tools, is capable of formalizing the uncertainties in the dynamic marketplace, and making service selection policies to achieve the best trade-off between costs and benefits. We use the causal-mapping approach to construct the structure of influence diagrams, and a Gauss kernel estimation method to estimate the marginal distributions of QoSs. Our experiments are based on simulated scenarios and real datasets. The experimental results show that the proposed framework is capable of accurately capturing the features of QoS values, predicting the QoS performance, and efficiently adapting to the changes in a long-term service consumption to facilitate policy determinations in a dynamic marketplace.