چكيده به لاتين
Auto-scaling systems can reconfigure cloud-based services and applications through various configurations regarding the cloud software layer as well as hardware provisioning. This will enable such systems to adapt to the changing environment at runtime. Such systems establish the foundation for achieving elasticity in the modern cloud computing paradigm. Given the dynamic and uncertain nature of the shared cloud infrastructure, cloud auto-scaling systems have been engineered as one of the most complex, sophisticated and intelligent artifacts created by engineers aiming to achieve self-aware, self-adaptive and dependable runtime scaling. In order to find an efficient solution to this problem, we need to be able to accurately predict the amount of workload and system metrics for future periods. Various solutions have already been devised to solve this problem. For example, many solutions make use of machine learning, statistical methods for time series and ensemble methods. In this research project, we examine the various strategies used to solve the problem of auto-scaling in cloud computing systems. Based on the understanding of the strengths and weaknesses of the existing researches, focusing on proactive mechanisms, we propose an approach for tackling this challenge. In this study, we see the problem as a sequence model, and we use convolutional neural networks to predict the workload of cloud services. Also, using neural networks, we obtain a mapping of predicted workloads and the real-time amount of needed resources to the future amount of required resources. In the last section, we propose a decision-making mechanism that takes into account different and sometimes conflicting criteria for the decision-making process and makes a compromise between the criteria. In the evaluation section, we examined the amount of prediction error, the amount of service level agreement violations, as well as the amount of resources’ under-utilization. Evaluations show that the proposed method has acceptable performance and accuracy. In addition, the proposed method for predicting the workload has shown a 4 percent improvement over the previous works.
