Using Virtual Machine Size Recommendation Algorithms to Reduce Cloud Cost

Abstract

Cloud spending has risen on a year-to-year basis, with the pandemic acting as the primary catalyst for its recent growth; however, “cloud waste,” referring to cloud resources that are not used to their full capacity, also follows this upward trend and causes the loss of an increasingly large amount of money. Unfortunately, present-day cloud research lacks data-driven studies that analyze why cloud users are wasting resources, or suggestions to users on how to lessen such waste. In order to prevent this over-expenditure, it is vital to choose the best-suited options when it comes to virtual machines (VM), especially for small to mid-sized businesses with limited funds and a lack of expertise. In this paper, we first analyzed the 235 GB Azure user dataset from the users’ perspective. We then implemented machine learning to determine our pricing model and the VM costs. With these statistics, we then delineated our methodology to calculate the wasted cost of each VM, and using this data, we propose an algorithm that can identify potential candidates with wasteful VMs and assist users in reducing costs. By applying our algorithm to approximately 2.7 million VMs, we demonstrate that it has the ability to help 66,721 VMs created by 1,520 users lower their monthly costs by $14.9 million. We conclude that businesses, while still reaping the benefits of cloud services, can do so at a much lighter cost and save on their VMs.