Introducing an adaptive model for auto-scaling cloud computing based on workload classification

With the increasing expansion of cloud computing services, one of the main goals of researchers is to solve its major challenges. Cloud service providers must satisfy the service level agreement for customers and prevent resource wastage as much as possible. Without a precise, optimal, and dynamic policy, this is unattainable. The key idea is the ability to acquire resources as you need them and release resources when you no longer need them, named "Cloud Elasticity." Elasticity is a trade-off between resource acquisition and release, and if this optimization is done best, the service level agreement will be fully achieved and the cloud provider will have the least waste of resources. The researchers used machine learning techniques to predict user workload and decide to scale up/out the resources. A challenging issue is the different characteristics of the users' workloads. The results show that each prediction algorithm works well on a class of users' workloads not all. Hence, in this study, a new architecture has been suggested to automatically classify the workloads based on their sequential statistical characteristics. First, the sequential statistical characteristics of the users' workload are extracted and then a trained neural network classifies the user's workload. The developed adaptive model chooses the best suitable algorithm among LR, SVM, and ARIMA to predict the workload. The results indicate a 10% improvement in forecast error.