Abstract:The precise estimation of resource usage is a complex and challenging issue due to the high variability and dimensionality of heterogeneous service types and dynamic workloads. Over the last few years, the prediction of resource usage and traffic has received ample attention from the research community. Many machine learning-based workload forecasting models have been developed by exploiting their computational power and learning capabilities. This paper presents the first systematic survey cum performance analysis-based comparative study of diversified machine learning-driven cloud workload prediction models. The discussion initiates with the significance of predictive resource management followed by a schematic description, operational design, motivation, and challenges concerning these workload prediction models. Classification and taxonomy of different prediction approaches into five distinct categories are presented focusing on the theoretical concepts and mathematical functioning of the existing state-of-the-art workload prediction methods. The most prominent prediction approaches belonging to a distinct class of machine learning models are thoroughly surveyed and compared. All five classified machine learning-based workload prediction models are implemented on a common platform for systematic investigation and comparison using three distinct benchmark cloud workload traces via experimental analysis. The essential key performance indicators of state-of-the-art approaches are evaluated for comparison and the paper is concluded by discussing the trade-offs and notable remarks.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the accurate estimation of resource usage in the cloud environment. Specifically, due to the heterogeneity of service types and the high - level dynamics of workloads, accurately predicting resource usage has become very complex and challenging. This high variability has led to over - or under - utilization of physical machines, as well as issues such as Service - Level Agreement (SLA) violations. To solve these problems, researchers have developed many machine - learning - based workload prediction models to utilize their computing and learning capabilities to improve prediction accuracy. Through systematic investigation and performance analysis, this paper compares multiple different types of machine - learning - based cloud workload prediction models, aiming to provide a comprehensive review and comparative study. ### Specific Problem Description 1. **High Variability of Resource Usage**: Resource requirements in the cloud environment vary greatly over time, leading to over - or under - utilization of physical resources. 2. **SLA Violation Problem**: During peak periods, virtual machine (VM) resource requirements exceed the available capacity of the server, resulting in problems such as performance degradation and increased response time. 3. **Resource Waste**: During low - demand periods, computing resources are wasted. 4. **Dynamic Adaptability**: The cloud environment is highly dynamic and requires prediction models to be able to quickly adapt to new workload patterns. 5. **Data Granularity and Prediction Window Size**: Selecting an appropriate data sample size and prediction window length directly affects the learning effect of the model. ### Solutions This paper proposes the following solutions: - **Systematic Investigation and Performance Analysis**: Classify and conduct systematic performance evaluations of existing machine - learning - based workload prediction models. - **Classification of Five Types of Models**: Divide different prediction methods into five categories: evolutionary neural networks, deep learning, hybrid learning, ensemble learning, and quantum learning, and discuss in detail the theoretical concepts and mathematical principles of each method. - **Experimental Platform Implementation**: Implement all five types of prediction models on a unified platform and conduct experimental analysis and comparison through three different benchmark cloud workload - tracking data sets. - **Key Performance Indicator Evaluation**: Evaluate the key performance indicators of existing state - of - the - art methods and discuss the trade - offs and significant features among various models. Through these methods, this paper aims to provide a comprehensive research framework for workload prediction in cloud resource management and provide directions and references for future research. ### Formulas Involved During the discussion, some formulas for data pre - processing and model training are involved. For example, the formula for normalizing data is as follows: \[ \hat{\omega}_{in} = x_a+\frac{d_i - \omega_{in}^{min}}{\omega_{in}^{max}-\omega_{in}^{min}}\times(x_b) \] where: - \( \hat{\omega}_{in} \) is the normalized input data vector. - \( x_a \) and \( x_b \) are the lower and upper limits of the normalization range respectively. - \( d_i \) is the original data value. - \( \omega_{in}^{min} \) and \( \omega_{in}^{max} \) are the minimum and maximum values of the input data set respectively. In addition, during the forward propagation process, the output calculation formula of the neural network is as follows: \[ \odot = D^\dagger\cdot W^\dagger + B \] where: - \( \odot \) is the result of the linear function calculation. - \( D^\dagger \) is the input data vector. - \( W^\dagger \) is the weight matrix. - \( B \) is the bias vector. These formulas ensure the correct processing of data and the effective training of the model.

Performance Analysis of Machine Learning Centered Workload Prediction Models for Cloud

A Comparison of Machine Learning Algorithms for Automatic Cloud Resource Scaling on a Multi-Tenant Platform

Workload Forecasting and Energy State Estimation in Cloud Data Centres: ML-centric Approach

Efficient resource utilization using multi-step-ahead workload prediction technique in cloud

Predictive Resource Allocation Strategies for Cloud Computing Environments Using Machine Learning

Experimental Analysis and Comparison of Load Prediction Algorithms in Cloud Data Center

Long Short Term Memory Recurrent Neural Network (LSTM-RNN) Based Workload Forecasting Model For Cloud Datacenters

Application-Oriented Cloud Workload Prediction: A Survey and New Perspectives

Performance prediction of parallel computing models to analyze cloud-based big data applications

CloudProphet: A Machine Learning-Based Performance Prediction for Public Clouds

Comparative analysis of cloud resources forecasting using deep learning techniques based on VM workload traces

Survey on prediction models in cloud resource management schemes

Cloud failure prediction based on traditional machine learning and deep learning

A deep learning-based resource usage prediction model for resource provisioning in an autonomic cloud computing environment

A hybrid cloud load balancing and host utilization prediction method using deep learning and optimization techniques

Workload Failure Prediction for Data Centers

A feature extraction and time warping based neural expansion architecture for cloud resource usage forecasting

Machine learning (ML)-centric resource management in cloud computing: A review and future directions

Forecasting Workload in Cloud Computing: Towards Uncertainty-Aware Predictions and Transfer Learning

Optimization of the Use of Cloud Computing Resources Using Exploratory Data Analysis and Machine Learning

Workload Forecasting Framework for Applications in Cloud