Budget in the Cloud: Analyzing Cost and Recommending Virtual Machine Workload

Abstract

Due to the increasing popularity of cloud computing, the cost companies pay to use the cloud and its services are growing significantly. Higher costs of cloud computing technology contribute to the volatility of the cloud and the financial instability of companies that depend on the cloud. Companies have their own unique cloud budget, and staying within that budget can become problematic in light of increasing costs. To save cost, many cloud users look to their cloud provider rather than looking at their own cloud purchases. A lack of studies on cloud users’ virtual machine (VM) workload and how cloud users spend money necessitates analysis of the spending of cloud customers. In this research paper, we analyzed a real-world data set from Microsoft Azure collected in 2019 that includes approximately 2.7 million VM traces. We developed a linear regression based pricing model to calculate the cost and used this model to analyze Microsoft Azure’s VM workload by comparing the cores, memory, lifetime, average utilization, and cost of each trace. By analyzing Microsoft’s data, we observed that users are not fully utilizing the cloud resources they have paid for. With this idea in mind, we then quantified the waste and developed an algorithm to determine which VMs are the most ineffective. We applied our algorithm to Microsoft Azure’s data set, and our results show that our algorithm discovered over one million wasteful VMs and helped 6,600 users save $3.3 million dollars. Even though cloud computing prices are increasing, cloud customers can save significantly by understanding VM workload better and selecting better-fitting VMs.