Downloads: 7 | Views: 278 | Weekly Hits: ⮙4 | Monthly Hits: ⮙7
Masters Thesis | Decision Science | India | Volume 12 Issue 8, August 2023 | Rating: 6.3 / 10
Enhancing Cloud Cost Efficiency: A Predictive ML Approach for Optimized Resource Allocation based on Infrastructure Usage and Workload Behavior
Apurba Kumar Roy
Abstract: There has been tremendous growth and development across the globe with respect to cloud adoption and usage. The traditional on-premises servers have given way to global cloud infrastructure with most of the vendors like Amazon, Microsoft and Google setting up dedicated cloud infrastructure and management provisions. These developments have allowed companies and organizations to concentrate more on building software applications and systems while the infrastructure part is totally taken care by the cloud providers themselves. While there has been rapid cloud growth and adoption across the industry, it has come with its share of cons too. Organizations worldwide started moving rapidly to cloud systems, migrating data and application, and paying only for what they used in terms of VMs, Clusters, CPUs, Memory, and components with zero maintenance cost associated with the infrastructure. But slowly few organizations began to realize that the cost that they started to pay for the usage of the Cloud infrastructure was becoming even more than what they used to pay for the traditional on-premises systems. Cloud adoption comes with numerous advantages, but when the resources and cost is not managed properly, the cost incurred can become huge and uncontrollable. This phenomenon requires an in-depth understanding and implementation of optimized resource management which would produce an optimized cloud cost resourcing model exploiting the workload characteristics and machine learning approach to produce substantial results. Using the real time Microsoft Azure workloads, we can predict the accurate optimized costing for future workloads which would help to take informed decisions towards cloud resource management and costing. The research proposes to take a closer look at the various aspects/Characteristics of Microsoft Azure Cloud implementation through the production Virtual Machine workloads and how we can take into consideration the cost perspective to take corrective steps to maintain a balance between the cost towards infrastructure and performance, at the same time reap the benefits of the cloud infrastructure in a more efficient and cost-effective way. The outcome of the research is an algorithm which considers all factors into account including uncertainty of workload behavior , different scaling options for different workload characteristics ,parallel running tasks factor and impact on workload behavior, adhering to the expected SLA and performance characteristic, efficient use of resource and avoiding resource wastage and higher cost expenditure, volume discounts for larger workloads, running times of VM having both short and long duration and predicting the future behavior based on the current selected workload thereby allowing provisioning strategies and pricing strategies from Azure costing portal. The research starts off with an analysis of the Azure workload and it characteristic followed by cleanup of the dataset. Thereafter the research intends to run various ML and Deep learning-based algorithms to ascertain the best features and co-relations among them to finally arrive at the best fit algorithm which considers all the factors and accurately predicts the cost & resource optimized model.
Keywords: Cloud Cost Efficiency, Machine Learning, Resource Allocation
Edition: Volume 12 Issue 8, August 2023,
Pages: 1590 - 1641