A S-Threshold Method to Perform Horizontal Auto-Scaling in a Cloud Computing Environment

Abstract

Cloud computing provides numerous benefits to individuals and any organization, making it a vital technology today. Users can communicate with the cloud computing systems and access the resources through the network. The need for resources can be increased, or at some time, it can be minimized according to the necessity of the resources. However, dynamic scaling is the pivotal property of cloud computing systems to dynamically devote the resources and manage the load while maintaining the quality of service requirement. This paper aims to provide a horizontal auto-scaling method that makes the system fault-tolerant, meets processing demand, optimizes the overall performance, meets user satisfaction, and ensures the quality of service requirement. The proposed S-Threshold method combines the State–action–reward–state–action (SARSA) method with the threshold model. It overcomes the drawbacks of the threshold method, automatically allocates the resources by adding or removing the machine according to the requisite basis, and ensures the aforementioned objectives. In this paper, the system initially relies on threshold-based rules for quick decisions. Over time, SARSA learns from experience and improves decisions, making scaling more adaptive and optimized. The proposed method juxtaposed with some existing methods and evaluated the five performance metrics to state the effectiveness of the proposed method. CloudSimPlus simulator simulates the proposed method into three significant categories: with various numbers of cloudlets, processing speeds, and VMs. While considering the increasing number of cloudlets, increasing processing speed, and increase in VMs, it noted that evaluated results of the proposed method give 50%, 80%, 85%, 60%, and 62% better results with respect to makespan, response time, waiting duration, average turnaround time, and throughput respectively.