Cloud Instance Selection Using Parallel K-Means and AHP

Abstract

Managing cloud spend and qualities when selecting cloud instances is cited as one of the timely research challenges in cloud computing. Cloud service consumers are often confronted by too many options and selection is challenging. This is because instance provision can be difficult to comprehend for an average technical user and tactics of cloud provider are far from being transparent biasing the selection. This paper proposes a novel cloud instance selection framework for finding the optimal IaaS purchase strategy for a VARD application in Amazon EC2. Analytical Hierarchy Process (AHP) and parallel K-Means Clustering algorithm are used and combined in Cloud Instance Selection environments. It allows cloud users to get the recommendation about cloud instance types and job submission periods based on requirements such as CPU, RAM, and resource utilisation. The system leverages AHP to select cloud instance type. Besides, AHP results are used by the parallel K-Means clustering model to find the best execution time for a given day according to the user's requirements. Finally, we provide an example to demonstrate the applicability of the approach. Experiments indicate that our approach achieves better results than ad-hoc and cost-driven approaches.