Publication

Abstract : Efficient mapping of tasks to Virtual Machines is significant in optimizing resource utilization and performance in Virtual Data Centers. In this work, a novel approach for mapping tasks to Virtual Machines by leveraging a bespoke probabilistic graphical model is proposed. The proposed framework maps tasks to Virtual Machines in two phases: constructing a Dynamic Bayesian Network to model scheduling factors and solving the mapping as an inference problem using the Junction Tree algorithm. The model updates the states of Virtual and Physical Machines during mapping, allowing for optimal mapping decisions based on evolving conditions. Experimental findings demonstrate that the proposed model outperforms the state-of-the-art models across various counts, achieving a makespan of 18.87 s, a degree of load imbalance of 0.22, a task guarantee ratio of 83%, an average Service Level Agreement violation rate of 0.32, and an energy consumption of 67 Kilo Joules. Compared to the strongest meta-heuristic baseline, the proposed model reduces makespan by 6.89%, decreases load imbalance by 11.08%, lowers Service Level Agreement violations by 12.44%, reduces energy consumption by 2.50%, and improves the task guarantee ratio by 4.28%.