Economic model based long-term IaaS service composition

The aim of this research is to develop an efficient and long-term cloud service composition framework from the Infrastructure-as-a-Service (IaaS) provider's perspective. First, a new quantitative economic model is developed that maximises the provider's long-term revenue and profit by selecting an optimal set of IaaS service requests in a dynamic economic environment. We propose a new multivariate Hidden Markov and Autoregressive Integrated Moving Average (HMM-ARIMA) model to predict various patterns of runtime resource utilisation. A heuristic-based Integer Linear Programming (ILP) optimisation approach is proposed to maximise the runtime resource utilisation. We deploy a Dynamic Bayesian Network (DBN) to model the dynamic pricing and long-term operation cost. A new Hybrid Adaptive Genetic Algorithm (HAGA) is developed that optimises a non-linear profit function periodically to address the stochastic arrival of requests. Next, we propose the Temporal Conditional Preference Network (TempCP-Net) as the qualitative economic model to represent the high-level IaaS business strategies. The temporal qualitative preferences are indexed in a multidimensional k-d tree to efficiently compute the preference ranking in runtime. A three-dimensional Q-learning approach is proposed to find an optimal qualitative composition using statistical analysis on historical request patterns. Finally, we propose a new multivariate approach to predict future Quality of Service (QoS) performances of peer service providers to efficiently configure a TempCP-Net. We have evaluated the efficiency of the proposed framework using Google Cluster data, real world QoS data and synthetic data. Experimental results show that the proposed composition framework efficiently maximises the provider's long-term economic goals in runtime. This research is expected to play a significant role in creating an economically viable and stable cloud market.