Cloud Workload Prediction: A Comprehensive Review of Techniques, Trends, and Open Challenges

Abstract

It is essential to predict workloads in cloud computing to enhance the scalability, performance, and cost effectiveness of cloud service offerings. As use cases continue to become more complex and volatile, it is essential to predict workloads as a step toward greater resource allocation and therefore less vigilance to maintain the predictability of the service. This review provides a synthesis and discussion of recent works grounded in workload prediction through traditional statistical approaches, machine learning, deep learning, and hybrid methods that combine methods. Each method will be assessed for their implications for the methodology as well as implications of the evaluation and prediction metrics including several benchmark dataset and evaluation measures. We will recognize and address pressing problems in workload prediction including the constraints de- scribed in a real-time prediction for a given dataset, and the challenges of a generalizable conclusions. Finally, we will identify opportunities for continued research and offer recommendations for future work to develop more adaptive, intelligent, and resilient workload prediction algorithms.